Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
& ENGINEERING
Volume No.14, June 2019
Innovation in the
Power Systems
industry
Enginers and specialists worldwide exchange
information and state-of-the-art world practices
to enhance knowledge related to power systems in
CIGRE’s latest publication.
Editorial Committee
Chief Editor Konstantin O. PAPAILIOU
Editors Pierre ARGAUT Erli FIGUEIREDO
Christian FRANCK Susana Almeida de GRAAFF
Nikos HATZIARGYRIOU Xidong LIANG
Pouyan POURBEIK Rob STEPHEN
Copyright registration N° 014, June 2019
1286-1146 CIGRÉ
© Copyright CIGRE
Conseil International des Grands Réseaux Electriques
International Council on Large Electric Systems
21 rue d’Artois - 75008 Paris - France
Tel : 33 (0)1 53 89 12 90
http://www.cigre.org
"The author(s) of papers published in the CIGRE Science and Engineering Journal, are
granted permission to post a PDF copy of their paper, as published in the journal, for self-
archival purposes on their own personal websites, or similar outlets (e.g. researchgate.com).
This permission is granted with the following restrictions, requirements, and understanding:
1. CIGRE is the copyright owner of all articles. Please see the CIGRE copyright statement for
details. The articles cannot be disseminate for commercial use (i.e. for payment, etc.).
2. A clear reference must appear where it is posted stating that "The paper may also be
downloaded for free from the source CIGRE. Goto: http://www.cigre.org/Menu-links/
Publications/CIGRE-Science-Engineering. Then click to download the <Month/Year> Issue
of the CIGRE Science and Engineering Journal. The paper appears on pages <page numbers>.
3. The paper can only be posted on the author(s) website after it has been published and
released through the CIGRE website."
With the longest day of the year (21st June) ante portas (in the northern hemisphere where I live) and
a prolonged period of nice sunny weather, I have to admit it is difficult to concentrate on reading the
12 excellent articles in the present issue of CIGRE Science & Engineering. The chances for our readers in
the southern hemisphere are much better since there, the longest night of the year is soon marching it
And believe you me it is more than worth! The present issue namely is a very nice mosaic of papers
displaying impressively the variety of knowledge sharing which makes CIGRE so unique.
To start with this issue contains two top Study Committee papers presenting the state of the art
on “Power system restoration – World practices & Future trends” from SC C2 and “The use of
battery energy storage systems for system integrity protection schemes in South Australian Power
System” from SC C4.
The second group contains five innovative papers on rather scarce issues such as “Wire
displacement of AAAC overhead line conductors at wedge tension clamp” and “Seismic response of
high-voltage cables in ducts – A preliminary study” as well as papers on “An aggregate dynamic model
for distributed energy resources for power system stability studies”, “A novel Wavelet selection
scheme for partial discharge signal detection under low SNR condition” and “Real-time co-
simulation model using electromagnetic transient and dynamic phasor simulations”.
It is followed by two papers, “IDtools: An automated tool for modal identification from time-domain
simulation results for establishing system operating limits” and “A direct calculation of locational
marginal value of distributed energy resources”, which have been selected as excellent papers from the
“The Grid of the Future” Symposium, sponsored by the CIGRE US National Committee (USNC) and
the Electric Power Research Institute (EPRI), and with the theme “Technology for the 21st Century
electric utility”, had been very successfully held on October 28 - 31, 2018 in Reston, Virginia, USA.
And last but not least, this issue includes the Grand Prize award - for authors under 35 years old
- for “Effect of Earth Surface on lightning electromagnetic pulse propagation” and the two
outstanding performance awards, i.e. “Material and emerging test techniques general overview of
AC and DC current injection on high voltage potential for HVDC long-term tests” and
“Application strategies for externally gapped line arresters against lightning outages on the 400 kV
overhead transmission lines in Hong Kong”, all three papers presented at the CIGRE-IEC 2019
Conference on EHV and UHV (AC & DC), which was held in Hakodate, Japan from April 23rd to
26th and which I had the honor to chair. The Hakodate Conference has been in my view one of the
highlights of this year with eight Study Committees and more than 330 participants from 27 countries,
which were more than impressed by the excellent organization and the unparalleled Japanese
hospitality. More details on this venue can be found in Electra No 304, June 2019.
If not enough, it is a remarkable coincidence, that Susana is the coordinator and corresponding author
of the SC C2 paper featuring on the top of this issue.
page
23
38
49
64
73
86
96
105
111
118
125
Restoration of the power system following an interruption Overview of past major incidents
is an important aspect of a System Operator’s (SO) role in
Depending on its operating point and defined constraints,
managing the bulk power system. Electric power grids in
a power system finds itself in one of the predefined
developed economies generally exhibit a very high degree
states. Figure 1 [1] depicts possible operating states of the
of reliability thanks to the well-established standards
power system, including different types of controls and
and criteria for the design, planning, construction and
contingency effects that facilitate the transition between
operations of the integrated network and the close
these states (figure originally from [1]).
interconnection in certain continents or regions. Despite
prudent planning and operations, major interruptions to the These states depend on the operational characteristics of
electric power grid (complete or partial blackout) do occur the power system and each jurisdiction can have its own
from time to time due to events (disturbances) that either definition of these states. For example, the European
exceed the basic design criteria, or due to various causes Network of Transmission System Operators for Electricity
such as natural disasters, multiple equipment failure, (ENTSO-E) defined these states as follows [2]:
protection relay miscoordination or malfunctioning, and • Normal State: means a situation in which the system
human errors. is within operational security limits in the N-situation
and after the occurrence of any contingency from the
When such disturbances occur, the power system may contingency list, taking into account the effect of the
experience wide-area, regional or local area blackouts available remedial actions.
with or without damage to, or prolonged outages of, • Alert State: means the system state in which the
major facilities. Restoring the integrity of the electrical system is within operational security limits, but
grid and supply to end-use customers is of paramount a contingency from the contingency list has been
importance to reduce and minimise undue hazard to detected and in case of its occurrence the available
social welfare, public safety, infrastructure security, and remedial actions are not sufficient to keep the normal
business activities. state.
To help restore the power system after major disturbances, • Emergency State: means the system state in which
control centre’s operating personnel are trained and one or more operational security limits are violated.
provided with a set of guidelines and procedures, defining • Blackout state (also in extremis state): means the
the strategy, to placing top priority on restoring stable system state in which the operation of part or all of
operation of the power system with sufficient energized the transmission system is terminated.
corridors so that power supply to other generation • Restoration state: means the system state in which
facilities and end-use customers can be restored as the objective of all activities in the transmission
expeditiously as possible to minimise interruptions to system is to re-establish the system operation and
social life and businesses. The key to successful and maintain operational security after the blackout state
expeditious restoration thus depends to a great extent on or the emergency state.
the SO’s preparedness, which includes operator training, The objectives of power system restoration (PSR) are to
availability of guidelines and procedure documents, enable the power system to return to normal conditions
effective communication protocols, provision and (i.e. normal or alert operating state) securely and rapidly,
assurance of blackstart capability, verification of cranking minimize restoration time and associated losses, and
path’s sustainability, etc. diminish adverse impacts on society [3]. Despite having
* susana.de.graaff@tennet.eu
By emergency
control Due to
occurrence of
more severe
In Extremis State contingency
Due to ineffective
or unavailable
emergency control
By emergency control
Emergency State
the same objectives, a review conducted by the CIGRE the effectiveness and efficiency in system restoration [5].
C2.23 working group concluded that system restoration An overview of the past major incidents is given in Table 1.
preparedness may vary from one SO to another, or may
This paper presents the more commonly adopted
differ between interconnected systems due to differences
approaches for power system restoration and speculates
in system characteristics and/or market design/rules
on several innovative approaches to tackle specific
[4]. In addition, the lessons learnt from actual events
challenges encountered during system restoration.
suggest that there are challenges which would warrant the
exploration of modified or innovative ways to improve
Table 1 Overview of Past Major Incidents [5]
____________________
1- Cranking paths are transmission corridors that extend from blackstart generation units to the targeted facilities needing offsite power.
• Creation of a blackout in the island and revival of the • Energise an isolated transmission test path to the
island according to the bottom-up approach. synchronisation point;
• Resynchronisation of the island with the main grid. • Manually synchronise the test system to the main
The steps of such an exercise are illustrated in Figure 3. transmission system. This is achieved by the controlled
synchronisation facility on the circuit breaker of a
The restoration trainings and associated debriefing
220 kV double busbar coupler, where one busbar is
sessions involve personnel from load dispatch centres,
energised by the blackstarted interconnector, and the
generating stations and distribution utilities and are being
other busbar energised by the transmission system;
organised for increased preparedness. This approach has
• Ensure pre-test conditions are met to protect the
shown to be helpful in keeping all stakeholders prepared
Irish transmission system from short circuiting or
and in reducing the time during actual restorations.
significant power flows in either direction.
Ireland Pre-Test Conditions
The East West Interconnector (EWIC) is a Voltage Source
After EWIC has successfully blackstarted, it operates
Converter (VSC) HVDC submarine cable connecting
in Islanded Mode, where the interconnector controls
Ireland to Great Britain. The interconnector employs
the voltage and frequency of the energised island. This
HVDC Light technology, which provides blackstart
network operation mode should not transfer to Normal
capability on both grids. EWIC is contracted by EirGrid,
Mode (power transmission mode) until the short circuit
the Irish TSO, to provide blackstart capability to the
level of the system it is connected with reaches 1800
Irish transmission system, and is required to perform one
MVA. This short circuit level is the lowest specified short
successful TSO requested local blackstart test per calendar
circuit capacity, determined by a dynamic performance
year. As part of the agreement, the TSO is also entitled to
study during the design phase of the interconnector. This
request EWIC to participate in a power system restoration
level is higher than what can be typically expected on the
test up to every five years, where EWIC will blackstart to
Irish transmission system in May, so additional generators
either provide a conventional generator with start-up load,
were requested to be synchronised in advance of the test.
or energise an isolated subsystem and synchronise with
the main transmission system. To assist with regulating the main system frequency for the
synchronisation of the test system and main system, it was
EWIC is one of two top-down blackstart sources for
agreed that the system should be synchronised after the
EirGrid; the other being the AC interconnection to SONI
morning demand peak had been reached at approximately
(Northern Ireland TSO). Bottom-up blackstart capability
10.00 hrs on the day of the test.
is provided by hydro, pumped storage and open cycle gas
turbine generators. Sequence of Events
1. HV switching to isolate a transmission test path from
Irish Restoration Test using VSC HVDC
HVDC converter station --> 400 kV transmission
In May 2018, EirGrid agreed a test plan with EIDAC
station --> 220 kV transmission station
(owner of EWIC) and the Distribution System Operator
2. AC supply to the converter station in Ireland is
(ESB Networks) with the following objectives:
disconnected
• Blackstart the interconnector locally from the 3. The low frequency trip level is artificially simulated
converter station blackout configuration; in the DC converter station
emergency conditions, assuring at the same time the effects were investigated by unbalanced load-flow and
absence of impact on final users. frequency-domain impedance analysis.
Before carrying out a real restoration test, it is necessary The electromechanical analyses are divided into different
to perform detailed analyses. The use of offline steady- steps. The first step is the collection of data. The Power
state and dynamic simulators of power network behaviour Network Simulator contains complete models of the
proves itself to be crucial for verifying the effectiveness Italian and part of the French transmission system,
of current and future plans. Electromechanical and including dynamic models of the generators (including
electromagnetic analysis are equally important in order capability curves, frequency and voltage regulation,
to analyse the performance of machines (including their power system stabilizer, real load ramp behaviour and
controls) on the one hand, and on the other hand to check synchronization devices) on the cranking path. For other
for possible critical effects due to for instance in-rush generators in Italy and neighbouring systems that are not
currents, mutual couplings or resonance effects. Such involved in the restoration test, the simulator contains
simulations require detailed dynamic models and careful simplified models.
parameter settings. This work is carried out through
The second step is to verify, together with the involved
data sheet analysis and real measurements processing,
stakeholders, the feasibility of a real test. The third step
respectively before and after the actual restoration test on
is the selection of the final test configuration. In this step
field.
certain protection settings are altered in order to avoid
Italy-France Restoration Test situations that can prevent the system from returning to
In 2012 Terna and RTE conducted the first European normal operation.
top-down restoration real test on a path of about 940 km
The full cranking path is modelled with the representation
starting from France (France to Valmontone, shown in
of Marti’s model lines (frequency dependent line model),
Figure 4. In about 25 minutes the total path was energized,
of different tower line types, including mutual coupling
considering the synchronisation of 3 hydro units near the
effects. Transmission lines that are not part of the cranking
French electrical area, with a total of 5 switching steps
path, but cause coupling effects with the cranking path are
implemented by Terna.
also modelled. The tower type has been represented by
Electromagnetic analyses were conducted in advance the geometrical parameters and physical characteristics
to verify possible critical aspects such as transient over- of phase and earth conductors (section, material, bundle
voltages, circuit breaker transient recovery voltage and spacing, etc.) and of the ground (soil conductivity).
rate of rise of restricting voltage values in case of fault Attention has been given to phasing sequence
opening during the energization process. Also resonance representation so that transposition can be considered.
237
236
10:57 - 10: 59:
235 REMESHING ACTIONS
234
233
232
10:41 a.m.:
231 SYNCHRONIZATION OF
PRESENZANO GROUP
230 10:31 - 10:32: END OF THE TEST
229 VARIATION OF TAP
CHANGER LAVORGO
228 TRANSFORMER 10:34 a.m.: CIRCUIT
BREAKER SUVERETO-
227 MONTALTO LINE CLOSED
226 10:22 a.m.: CIRCUIT BREAKER 10:51 a.m.: CIRCUIT
MUSIGNANO-LAVORGO LINE CLOSED BREAKER SUVERETO-
225
STARING POINT OF THE TEST MONTALTO LINE OPENED
224
Restoration training goals, methods, content and • Principles of power system stability controls
organisation • Identification of possible emergency states
Operator training is a crucial element for the successful • Operation of the power system in the emergency state
execution of designed restoration plans. Figure 6 [9] • Handling power shortage situations (i.e. generation
illustrates the frequently used training goals. The training adequacy)
of the operators covers several operating conditions, • Principles of load shedding
including blackstart and restoration procedures. • Power system restoration principles and procedures
The training itself can be done in different ways, from The main tool for restoration training today is an operator
lectures over OTS to innovative E-learnings. Lectures, training simulator (OTS), also known as dispatcher
site visits and on the job training are done by almost training simulator (DTS). The survey conducted in [10]
all companies surveyed in [10], while OTS training is identified that 80%, 90% and 90% of those with an OTS
performed with slightly lower frequency as shown in use it for training of conditions with normal power system
Figure 7 [9]. state, for emergency handling and system restoration,
respectively. After the major incidents that occurred
E-Learning is an upcoming training method, owing this
worldwide in the last 10-15 years, we can expect that
to its flexibility with respect to time and place. All these
practically all companies that do use an OTS/DTS for
trainings methods have advantages and disadvantages,
operators’ training, use it extensively for emergency
therefore a mixture of different methods can provide a
handling and restoration training.
good balance.
An example of an OTS/DTS architecture, including
For effectively handling the emergency and restoration
subsystems Power System Model (PSM), Control Centre
power system operation states, an advanced set of
Model (CCM) and Instructional Subsystem (ISS), is
training modules is needed to raise the operator’s
illustrated in Figure 8 [10].
knowledge and skills. Such a training module would
contain, for example: The PSM is responsible for the realistic representation
• Principles of power system dynamics of all basic power system elements (generation, network
and consumers, together with their main control deteriorate as the system synchronous inertia decreases,
and protection devices). Apart from modelling, this the short-circuit power levels across the network will
subsystem also includes algorithms that simulate all become more volatile, and there will be more scarce and
relevant dynamic behaviour of the power system. distributed voltage control resources. Currently, there
The CCM is responsible for the exact representation of is very limited provision of ancillary services, such as
the power system control centre equipment (typically voltage control or balancing, by RES. Furthermore, it is
SCADA/EMS system) that operators use in their daily still not common practice to redispatch (e.g. curtail) RES
work to monitor, analyse, support decision making and for congestion management. There is also no restoration
finally control the system. The operator training on a service capability yet in RES. It is also the case that TSOs
replica of their SCADA/EMS system yields the highest currently have little or no observability and controllability
possible training effect, though use of generic CCM is of RES – both factors being a pre-requisite for coordinated
possible too (mainly for improving understanding of the power system restoration.
power system reaction on the operator’s actions). This transition requires SOs to make a paradigm shift to a
Finally, the ISS should enable the instructor to monitor more diverse and decentralised set of resources to actually
and control (start, stop, pause) a training session, control and restore the power system, including vertical
introduce events, emulate other non-modelled parties (TSO/DSO) and horizontal (TSO/TSO) coordination,
and support creation, validation and maintenance of together with a new set of awareness and decision support
training scenarios. tools, which include the measurement and analysis of
system dynamics (using WAMS and DSA). Moreover,
In the simulator training environment the operators can
the increasing integration of power electronics into the
train under task-relevant stress conditions, i.e. conditions
power system creates new possibilities to effectively use
that characterize real life incidents/restorations should
the available technology.
also be recreated during the training session. Stress
conditions can for example be emulated with additional Impact of Transition to RES on Restoration
(numerous) phone calls, by shortening of some time The increasing penetration of RES, especially based on
constraints or resources, and/or by introduction of wind and solar energy, introduces a volatile production
unknown events. pattern in the generation mix due to its intermittent nature.
During system restoration, SOs have as common practice
3. Restoration in the Future to still use conventional power plants to restore the power
TSOs’ experience and knowledge are mainly based on a system, making the process stable and predictable. In a
“traditional” power system, where conventional generators future where less or no synchronous generators are going
have an essential role in maintaining the system’s security. to be available, it is important to rethink the restoration
It is now possible for system dynamics to be monitored strategies, using all technologies available in the power
in real-time by control room operators or analysed in system with the capability to provide support.
the operational planning timeframes, depending on the Due to the fact that the share of RES in distribution
specific power system (e.g. level of renewable energy networks is nowadays significant with the tendency to
sources (RES) penetration, size and stability conditions). grow, there is a need for TSO/DSO integrated restoration
In the future, the power system is expected to become plans, which will involve increased coordination,
more dynamic and less stable if no additional measures are information exchange, joint operator training, and most
taken [11], [12]. It is expected that frequency stability will likely common tools.
Therefore, increased visibility is required of the low of the shed load, resynchronisation of multiple islands
voltage network, as well as greater coordination, e.g. and automated building up of power generation. The
concerning voltage control. automated algorithm assists the operator in rebuilding the
system again with reduced time [18].
Storage can also be used as blackstart sources that provide
the required power to non-blackstart generators. In [17] In 2004, WAMS were used during the resynchronisation
for example, a case study of a hybrid system consisting process of the first and second UCTE zone, and again
of a BESS and a STATCOM is presented. This hybrid in 2010 during the connection phase of Turkey to the
system was able to provide energizing power of 27 MW Continental European system. By doing so an on-line
for 15 minutes, which was considered sufficient for initial dynamic assessment of the system stability during
restoration. the “hot” synchronisation phase was performed. The
policies, guidelines, as well as responsibilities during
WAMS support for system restoration
resynchronisation for the ENTSO-E jurisdiction are
Fast system dynamics resulting from the integration of
documented in Policy 5 of the Continental Europe
PEID adds complexity to power system operation. Wide
Area Monitoring Systems based on Phasor Measurement Operation Handbook [19].
Units (PMUs) are being used by System Operators In India, synchrophasors are being used in control
worldwide in the operational environment, giving the rooms for real time system monitoring and analysis
control room information about the dynamic behaviour [20], [21]. Since PMUs send time synchronised signals
of the network and, consequently increasing awareness of frequency/voltage phasors, the synchronisation relay
for system dynamics. Next to the improved situational available at substations could effectively be mimicked
awareness and decision support, the synchrophasor at control centres. This helps in coordination between
technology in the control room can definitely contribute to generating station(s) in different islands and synchronising
power system restoration. When compared to traditional substations for matching of electrical parameters and
SCADA measurements, synchrophasors have an added faster synchronisation of islands. The above is being used
value of synchronised voltage phase angle information during synchronisation of islands during mock blackstart
between areas that have to be re-energized and/or re- exercises and is likely to prove beneficial during actual
connected, which can significantly benefit the restoration restoration process.
process. In the preparation phase of the restoration
The Italian practice, as highlighted in Section 1 is another
process, and when complemented with state estimation
example of WAMS usage for restoration purposes.
data, the synchrophasors provide precise information of
the remaining system, its division in islands and available DSA support for system restoration
components in the system. This information helps to A list of enhancements worth considering to further
construct the restoration strategy. From a restoration improve the restoration effectiveness and efficiency
viewpoint, the restoration stage can be enhanced with is presented in [4]. One of them is adopting advanced
critical data such as synchrophasor measurements from SCADA and EMS functionalities, such as wide-
generating units and critical load. In the restorative area security assessment or a fully integrated control
control module (as part of the SCADA), the algorithm landscape including SCADA/EMS, PMU within Wide-
constituting of synchrophasor and state estimation data Area Control Systems and advanced analytics tools (e.g.
can help in the automated process of re-energization dynamic security analysis, DSA, and the like) to enhance
* nilesh.modi@aemo.com.au
KEYWORDS
Battery Energy Storage Systems, Special Protection Scheme, Loss of Synchronism
SA power system to recover to a stable operating state and 1. However, this analysis does not include the effects of
avoid a large frequency excursion. Operation of stage 3 this more recent addition. Further transmission connected
would separate the SA region from the rest of the NEM. batteries are expected to be included in the scheme as
However, to achieve stable operation after the activation they come online. Stage 1 utilises the 70 MW of BESS
of stage 3, the SA region needs to rely on UFLS to restore reserve from HPR, with the aim of avoiding LoS between
the supply-demand balance. the SA power system and the rest of the NEM. Active
2.1. Stage 1: Energy discharge from BESS power is rapidly injected from the BESS while SA is still
synchronously connected. The stage 1 SIPS activation
On 1 December 2017, the world’s largest grid-connected signal to the BESS is initiated if active power flow into
lithium-ion battery (often referred to as the “Tesla SA over the Heywood interconnector either:
Battery”), was commissioned. This BESS is officially • Increases at a rate which is faster than a given
known as the HPR and is located approximately 200 km threshold, or
north of Adelaide. It is rated for 100 MW / 129 MWh • Increases above a specified absolute value.
capacity. Of the 100 MW instantaneous capacity, 70 MW
is reserved for power system security purposes, with the Once the trigger is received, the BESS will immediately
remaining 30 MW available for commercial operation. inject 70 MW into the SA power system, giving time for
either generator frequency control mechanisms to respond
Operation of the BESS to date suggests that it can provide or if unavailable, for the UFLS to operate.
a range of valuable power system security services,
including rapid and accurate responses to frequency Figure 2 shows a simplified schematic diagram that
deviations and frequency control services [7]. The illustrates the logic on which SIPS stage 1 operates.
BESS has been configured to provide post-contingency This scheme was designed by ElectraNet who is the
frequency control support at all times. transmission asset owner in the SA.
Figure 1 shows a comparison between the response of As shown in Figure 2, the operation of SIPS Stage 1 is
a thermal generating unit and battery during an under- triggered by active power levels flowing into SA through
frequency and subsequent over-frequency event in the the Heywood interconnector. Based on operational
NEM in 2018. The thermal unit and battery injected experience, thresholds for both the level and the rate-of-
additional active power of approximately 45 MW and change of active power have been determined.
80 MW, respectively, before the frequency nadir. More
2.2. Stage 2: Pre-emptive load shedding
importantly, when the frequency turned around, the
battery immediately started absorbing power thereby The aim of stage 2 is to quickly detect power swings that
providing a greater contribution to arresting the frequency would otherwise lead to LoS if not rapidly managed.
peak than the thermal unit. The stage initiates pre-emptive load shedding (i.e. before
frequency reduces to the UFLS trigger levels) so as to
When this analysis was carried out, only the HPR was
avoid a LoS condition between the SA power system and
integrated into the SIPS stage 1. Recently another
the rest of the NEM.
transmission connected battery near Dalrymple substation,
known as ESCRI, was also integrated into the SIPS stage The activation signal is initiated from a LoS relay set
LOAD
SHEDDING SOUTH EAST
LOS
LOS SUBSTATION
TRIGGER TO
RELAY
RELAY REMOTE
TAILEM BEND SUBSTATIONS
SUBSTATION
Stage 2 Trigger
to look for power swings that would otherwise lead to the outer and inner blinders within a specified timeframe,
instability. This LoS relay is installed at a substation deeper then the relay operates. An example characteristic of each
within the SA power system than the first interconnector relay is shown in Figure 3.
substation, but still on the primary path for inter-regional
power transfers (Tailem Bend substation in Figure 2). X
As further explained in Section 2.4, this substation is
Location of LoS
also the natural electrical centre of the SA power system protection
Location of LoS
protection (South
and the power system of the neighbouring NEM region (Electrical Center of
East)
SA – Tailem Bend)
(Victoria). This means that phenomena associated with
LoS (i.e. depressed voltage and increased power flow) are
likely to manifest at this location before other substations
in the region.
R
The operation of this LoS relay is similar to that of the 2.5. System conditions, study methodology and
South East LoS relay. However, the intent of this relay is assessment criteria
to pre-emptively disconnect loads within the SA region. To assess the effectiveness of the BESS’s contribution
On detection of a power swing (i.e. loss of synchronism), to the overall SIPS, a wide range of operating conditions
it sends a trip signal to predetermined circuit breakers to were simulated in a region-wide Electromagnetic
disconnect load. Transient (EMT) model of SA. For each of the system
conditions, a set of contingency events were investigated
as listed in Table 1.
To assess the effectiveness of the BESS, an EMT model 3. Results and Discussion
of the SA region was developed. The model development
This section presents key findings from the analysis
methodology, details about the models and their integration,
undertaken. In this paper, only results associated with key
benchmarking of the models are available in [8]. As
the SA region has substantial asynchronous generation scenarios are discussed. Table 2 provides detail of those
compared to its demand, it was more appropriate to assess key scenarios.
the SIPS design using highly detailed EMT type models. 3.1. Loss of multiple asynchronous generators –
Once an EMT model of SA was established, a number of Scenario B1
scenarios to reflect various system operating conditions
In this scenario, 300 MW of asynchronous generation
were developed and analysed.
was tripped simultaneously. To investigate the ability of
The effectiveness of stage 1 (BESS) was assessed based each stage to defer the activation of the subsequent stage,
on the following criteria: AEMO studied the following sensitivities.
• Effectiveness of stage 1 in avoiding a stage 2 trigger.
3.1.1. All stages disabled
• Effectiveness of stage 1 in improving overall power
system security in SA. The intent of this sensitivity study was to confirm that a
multiple contingency event that leads to a total generation The following observations can be made regarding the
loss of 300 MW can trigger LoS between SA and the activation of stages 2 and 3 of the SIPS.
rest of the NEM. It is apparent from Figure 6 and Figure • Stage 2 of the SIPS (i.e. pre-emptive load shedding)
7 that such an event can trigger a LoS condition without correctly triggers and disconnects load.
mitigating action from SIPS. Figure 6 shows how the • The voltage phase angles stop diverging, as shown in
voltage phase angles between South East and the rest of SA Figure 8.
start to diverge, indicating an onset of LoS. This is further • The trigger of SIPS stage 3 is avoided. Figure 9
confirmed in Figure 7, which shows the impedance loci shows that the impedance locus of stage 3 does not
measured at both Tailem Bend and South East substations.
cross the inner blinder of the LoS relay.
3.1.2. Stage 1 disabled, Stages 2 and 3 enabled • The voltages throughout SA return to new steady-
The intent of this sensitivity study was to test the state values as shown in Figure 10.
effectiveness of the SIPS stage 2 pre-emptive load shedding • Figure 11 shows the amount of load that was
in avoiding the onset of stage 3 (i.e. disconnection of SA disconnected. On disconnection of this load, the
from the rest of the NEM) in the B1 scenario. Figure 8 SA power system regains synchronism with the rest
shows the voltage phase angles and re-confirms the onset of the NEM and starts towards a new steady-state
of LoS between SA and its neighbouring region. operating point.
3.1.3. All stages enabled trigger crosses the outer blinder, but does not cross
the inner blinder and trigger further SIPS action. The
The intent of this sensitivity study was to assess the
locus returns beyond the outer blinder towards a new
ability of SIPS stage 1 to avoid triggering stage 2 in the
steady state value.
B1 scenario. The process by which SIPS stage 1 acts is
shown in Figure 2. 3.2. Loss of multiple synchronous and asynchronous
The following observations can be made for this scenario: generators – Scenario E1
• Rapidly discharging BESS to the grid prevents the Scenarios with multiple synchronous and asynchronous
development of unstable power swings in the SA generation disconnections were considered. The following
system. Voltages and voltage phase angles across the is an example of the loss of 470 MW of synchronous and
region return to new steady state values relatively asynchronous generation.
quickly. This is shown in Figure 12 and Figure 14.
3.2.1. Stage 1 disabled, Stages 2 and 3 enabled
• Stage 1 (i.e. energy discharge from BESS) avoids the
trigger of Stage 2 (i.e. pre-emptive load shedding). The intent of this sensitivity study was to identify the
The impedance locus shown in Figure 13 confirms effectiveness of SIPS Stage 2 in deferring the activation of
the avoidance of a LoS condition. As shown in Stage 3. Figure 15 and Figure 16 show that in the absence
Figure 13, the impedance locus feeding the Stage 2 of SIPS Stage 1 action, these multiple contingencies are
KEYWORDS
Distributed Energy Resource Modeling
Bss
M Motor C
A/C Motor D –
represents 1-phase
residential a/c
Electronic
Load
Static
DG
Load
incorporated into the composite load model. In section 3, constitute well over one-hundred parameters and so are
a description is given of a detailed testing that has been perhaps too complex for modeling aggregated DER, and
done of the model across four commercial software (ii) the 2nd generation models were developed for modeling
platforms. Also, in section 3, a brief description is given single large wind/PV and battery-energy storage plants and
of running the model in large system studies, together so may not provide a simple means to represent aggregated
with the composite load model, to compare it with PVD1. behavior across numerous distributed generators.
Finally, section 4 provides the conclusions and summary
Thus, starting with the model structures of the 2nd
as well as some brief comments on future work.
generation RES models (i.e., repc_a + reec_a + regc_a),
a significantly reduced version of the core functionality
2. The DER_A Model was developed to form DER_A. Figure 2 shows the new
The ultimate purpose of the distributed energy resource DER_A model. The complete parameter list for the model
model version A (DER_A) is for it to be used to represent is given in Table 1. The model has 48 parameters and 10
the aggregated dynamic behavior of the DER in time- states, which is roughly 1/3 of the number of parameters
domain positive-sequence stability studies. That is, of the full large-scale 2nd generation RES generic models
this model would represent the combined (aggregated) [1]. Nonetheless, it preserves a significant number of those
dynamics behavior of many tens to hundreds of small features, namely frequency and voltage control emulation,
distributed inverter-based generators on the distribution with asymmetric deadband. The voltage control only
system on for example residential feeders – such as roof- allows for proportional control, while the model also
top photovoltaic generation. As such, it will eventually be allows for constant power factor and constant Q-control.
deployed as part of the composite load model, as shown It is possible, however, that both constant Q-control and
in Figure 1. proportional voltage control or constant power-factor (pf)
The concept behind development of the DER_A model was control and voltage control are in effect simultaneously.
to create a model that is able, to some extent, to emulate For example, assume the distributed generation is in
the key dynamic performance that may be required from constant Q-control, holding a small lagging power factor
such resources in the future, such as frequency and voltage (or in constant pf-control at a small lagging pf), such that
control. At first sight the 2nd generation generic renewable it is generating 0.1 MVAr on a 2 MVA unit. Then assume
energy source (RES) models [1], that were developed for that Kqv = 10 (proportional voltage control gain) and
inverter-based generation, may seem appropriate to use to dbd1=dbd2=0.05 with Vref0 = 1.0. Now so long as the
model DER (i.e., repc_a + reec_a + regc_a). However, voltage remains within 0.95 to 1.05 pu, the Q output of the
there are two drawbacks with this approach (i) they unit remains at 0.1 MVAr. If an event occurs to depress the
Iql1
Iqh1
Vref0 (user defined)
pfaref tan
Iqmax
+
Pgen 1 + Iq
1 1 1
(Pord)
1 + s Tp 1 + s Tiq 1 + s Tg
Iqcmd
0.01
s1 Qref s2 s3
0 Iqmin
Pflag
Vt_filt Xe
(s0)
Vrfrac
~
Ed = V td o � iq X e
Eq = V tq o + id X e
+
Q priority Pqflag Current
1 j V
Iqmax = Imax 0 – Q priority Limit 0
Iqmin = -Imax 1 – P priority Logic
1
1 +
Ipmax = sqrt( Imax2 – Iqcmd2) 1 + s Tv
Vt_filt 0 1
P priority (s0) vl0 vl1 vh1 vh0 s4 Vtripflag
Ipmax = Imax
Iqmax = sqrt( Imax2 – Ipcmd2)
Iqmin = -Iqmax
Vt_filt Frequency Tripping
(s0) Logic See Details in
0.01
Model Description Id ( = Ip)
Freq_flag
0 Pmax rrpwr
Pref dPmax Ipmax
Pref
Pmax Ip
s7 1 1
+ femax
_ 1 + s Tpord Pord Ipcmd 1 + s Tg
1 Kpg + Kig
Pgen 1 + s Tp dPmin s8 s9
s Ipmin
(Pord) 1 -rrpwr
+ Pmin
femin s6
0 Pmin
fdbd1,fdbd2 Ddn
Frq_filt +
_
1
Freq
1 + s Trf
s5 + +
Dup
Freq_ref 0
0
To frequency relay model
1 1
If Vt (terminal voltage) ≤ Vpr then switch to
position 1, else position 0
Freq_ref
power feedback from the power-order (Pord) prior to the WECC model specification document [7] but the most
Vrfrac block, this problem is avoided. This ensures that salient points are summarized here for completeness.
Pord is always equal to Pref, which is what is desired. The values of Vt and Pgen are the voltage and electrical
Furthermore, the user should be allowed to set Tpord and power at the terminals of the DER_A model. Upon
Tp to zero (0). By doing so and setting Kpg = 0 and using initialization, Pref and Qref will be determined in software
a non-zero value of Kig, a simple proportional only droop- to properly initialize the model. If Kqv is non-zero, then
control can be effected, since the closed loop around Pord upon initialization dbd1 < Vt - Vref0 < dbd2, where Vref0,
in this case creates a simple time-constant equal to 1/Kig. dbd1 and dbd2 are user defined value. If this condition
In this case, i.e. when Tpord = Tp = 0, Kig cannot be set to is not met, then the software tool will force Vref0 = Vt
zero, it must be a positive number. For similar reasons, the and indicate this to the user in a warning message. If
feedback to the power factor controller is also from Pord. dbd1=dbd2=0 (which should typically not be done, since
The frequency tripping is modeled in simple terms. If these distributed generation models are not intended to
frequency goes below fl for more than tfl seconds, then tightly control voltage) and Kqv is non-zero, then the
the entire model will trip. If frequency goes above fh for program should give a warning/error message to the user
more than tfh seconds, then the entire model will trip. This and indicate that Vref0 has been set to equal to Vt (to force
block is disabled, if voltage is below Vpr, to avoid tripping the error to zero and thus the output of the voltage leg
on frequency spikes (as calculated in simulation) due to to zero); the initial Q from power flow is then initialized
sudden voltage drops. This is depicted in Figure 2, and off of the constant Q/pf leg. This is the simplest solution
shown in more detail, in an expanded view, in Figure 4. in this case. Finally, during initialization, the software
program should check to ensure that the terminal voltage
The model may also be used to emulate inverter-interfaced (Vt) of the model initializes to a value that is greater than
distributed energy storage. This is achieved by allowing vl1. Also, vl1 must be greater than or equal to vl0. If either
the model to absorb, as well as generate, real power. The of these conditions are not met, the program will present
additional flag, typeflag, achieves this. When typeflag is set an error message to the user indicating that the value of
to 0, meaning the device is a generator, Ipmin (that is, the vl1 and vl0 are inappropriate, and thus the model will
minimum active current limit) is set to zero. When typeflag ignore the Vrfrac block. A similar check should be made
is set to 1, meaning the device is a storage device, Ipmin is on vh1 and vh0. Also, a check should be made to ensure
set to -Ipmax. Need-less-to-say, for the sake of simplicity, that tvl1, tvl0, tvh1and tvh0 are all greater than or equal to
there is no attempt to model the storage mechanism (e.g. zero. There is no limitation on which of these timer values
charging/discharging of a battery) and so it is assumed that should be greater or smaller. The Vrfrac block is explained
the model would only be used for transient simulations in more detail in the appendix.
(e.g. 10 to 30 seconds) during which there would be no
appreciable effect of the storage mechanism. There is a possible control problem. If this model were
used to model a single large inverter-based device
A simple representation of the voltage source interface connected to a weak grid point (i.e. low short-circuit ratio)
that is employed by most equipment vendors (based on where the voltage is highly affected by this device, then
[6]) is also modeled, since by far the majority of inverters there could be a possibility for limit-cycling (i.e. voltage
used for inverter-based resources are current-regulated
goes outside deadband, device brings voltage inside
voltage-source converters.
deadband by changing Q, Q drops to constant initial value
The details of initializing the model can be found in the once voltage is within the deadband, voltage goes outside
deadband, etc.). For this, and other reasons, it is in general, 3. Testing the DER_A Model in
not recommended that this model be used to model large
plants. Four Commercial Simulation
Two other important notes should be made. First, that the
Platforms
filtered value of voltage (Vt_filt) and frequency (Frq_filt) Once the model was defined and agreed to, four major
is used in all the controls and timers. Second, that the commercial software vendors in North America, decided
current limit is modeled as follows: to graciously implement the model in their respective
software tools so that it could be tested. This testing was
a. Q-priority: Iqmax = Imax; Iqmin = -Imax; done in two steps. First, a beta version of the model was
Ipmax = Imax 2 � Iqcmd 2 ; if typeflag = 0 then released by all the software vendors and a set of test
Ipmin = 0, else Ipmin = - Ipmax protocols were defined [8]. The tests aimed, to the extent
b. P-priority: Ipmax = Imax; Iqmax = possible, to test all the features of the model and ensure
Imax 2 � Ipcmd 2 ; Iqmin = -Iqmax; if typeflag = 0 that they performed as expected, as well as to benchmark
then Ipmin = 0, else Ipmin = - Ipmax the model across the four commercial platforms to ensure
consistent implementation and performance across all
A final note is that the post-fault rate of recovery on active- the software tools. A simple test case was developed
current (rrpwr) is also imposed (in the opposite direction) for performing simulations in all the various software
when the model is being used to “emulate” charging of platforms, it is shown in Figure 5. A complete list of
an energy storage device. That is, when Pgen is negative, all the tests and results may be found in a report by the
then rrpwr is applied with its sign changed and it becomes Electric Power Research Institute (EPRI) [8]. These tests
the ramp-rate at which charging power (power being were focused on testing the DER_A model as a standalone
absorbed by the model) increases after a fault. model. The testing proved successful and so the model
Figure 6: Plots from one of the benchmarking tests – Test 1A – voltage sag with a ramped recover.
a) Net Load in the local area (MW) b) Total distributed generation in the area (MW)
Figure 7: Simulation results form a large system model (WECC system) where the distributed generation is represented within the composite load model first using
the old PVD1 model and then using the new DER_A model.
was approved and released on all the software tools. and simulating some major system events in one region
Figure 6 shows one example of the many tests performed. with (i) all the distributed generation initially modeled
This testing was lead and performed by EPRI. During using the old PVD1 model [4], and (ii) by replacing all the
the testing one case was found to have some small, but PVD1 models in the composite load model with the newly
noticeable, difference in response among the tools, where developed DER_A model. Furthermore, some sensitivities
two of the tools matched and the others had a slight were performed on the model parameters. Example plots
difference. This was for the case of playing into the model from this work are shown in Figure 7. The conclusions
a frequency wave-form. Upon closer investigation, it that may be drawn from this analysis, as seen in Figure 7,
was identified that the differences were due to numerical are as follows:
precision of the integration schemes, and thus for now 1. The DER_A model seems to perform well in a large
this was not further investigated. Some of these subtle system model.
differences in the frequency calculation can actually be 2. If the parameters of the DER_A are properly adjusted,
seen in the frequency traces in Figure 6. it can be made to emulate the older, and much simpler,
PVD1 model – this can be seen by the fact that the
The next step in testing the model was to incorporate it brown and blue lines in the simulations (Figure 7)
into the composite load model (Figure 1) and to then test match for the total net load and distributed generation
it by using the model to simulate distributed generation in the area, which is driven by the performance of these
across a large system. This testing was done by one of models.
the task force members [9], by taking a WECC base case 3. Having the DER modeled as a part of the composite
Parameter Description
Trv transducer time constant (s) for voltage measurement
Trf transducer time constant (s) for frequency measurement (must be ≥ 0.02 s)
dbd1 lower voltage deadband ≤ 0 (pu)
dbd2 upper voltage deadband ≥ 0 (pu)
Kqv proportional voltage control gain (pu/pu)
Vref0 voltage reference set-point > 0 (pu)
Tp transducer time constant (s)
Tiq Q control time constant (s)
Ddn frequency control droop gain ≥ 0 (down-side) (pu/pu)
Dup frequency control droop gain ≥ 0 (up-side) (pu/pu)
fdbd1 lower frequency control deadband ≤ 0 (pu)
fdbd2 upper frequency control deadband ≥ 0 (pu)
femax frequency control maximum error ≥ 0 (pu)
femin frequency control minimum error ≤ 0 (pu)
Pmax Maximum power (pu)
Pmin Minimum power (pu)
dPmax Power ramp rate up > 0 (pu/s)
dPmin Power ramp rate down < 0 (pu/s)
Tpord Power order time constant (s)
Kpg active power control proportional gain (pu/pu)
Kig active power control integral gain (pu/pu/s)
Imax Maximum converter current (pu)
vl0 voltage break-point for low voltage cut-out of inverters (pu)
vl1 voltage break-point for low voltage cut-out of inverters (pu)
vh0 voltage break-point for high voltage cut-out of inverters (pu)
vh1 voltage break-point for high voltage cut-out of inverters (pu)
tvl0 timer for vl0 point (s)
tvl1 timer for vl1 point (s)
tvh0 timer for vh0 point (s)
tvh1 timer for vh1 point (s)
Vrfrac fraction of device that recovers after voltage comes back to within vl1 < V < vh1
fl frequency break-point for low frequency cut-out of inverters (pu)
fh frequency break-point for high frequency cut-out of inverters (pu)
tfl timer for fl (Tfl > Trf) (s)
tfh timer for fh (s)
Tg Current control time constant (s)
rrpwr Power rise ramp rate following a fault > 0 (pu/s)
Tv time constant on the output of the voltage/frequency cut-out (s)
Vpr voltage below which frequency tripping is disabled (pu)
Pflag 0 - for constant Q control, and 1 - constant power factor control
Pqflag 0 - Q priority, 1 - P priority for current limit
Freq_flag 0 - frequency control disabled, and 1 - frequency control enabled
Ftripflag 0 - frequency tripping disabled; 1 - frequency tripping enabled
Vtripflag 0 - voltage tripping disabled; 1 - voltage tripping enabled
typeflag 0 - the unit is a generator Ipmin = 0; 1 - the unit is a storage device and Ipmin = - Ipmax
Xe Source impedance reactive > 0 (pu)
Iqh1 Maximum limit of reactive current injection (pu)
Iql1 Minimum limit of reactive current injection (pu)
Irina Green received here BE and MSEE in electrical Songzhe Zhu received her BSEE from Xi’an Jiaotong
engineering from the Peter the Great St. Petersburg University, China in 1993 and MSEE from Nanjing
Polytechnic University, Russia. She has extensive Automation Research Institute, China in 1996, and Ph.D.
experience in transmission planning and power system from Iowa State University, USA in 2000. She is currently
analysis. She was a senior transmission planner with a Senior Advisor with Regional Transmission at California
Pacific Gas & Electric from 1997 to 1999. Since 1999 she ISO.
has been with the California ISO, and presently serves in Matthew Torgesen received his BS in Electrical
the role of Senior Advisor. In this role she helps to identify Engineering from Utah State University in 2013. He has
and resolve engineering-related problems within assigned worked for several key industrial companies in a power
areas of the CAISO-controlled grid, and extensively works engineering capacity. From 2014 to 2018 he was with the
on the integration of renewable resources and distributed Western Electricity Coordinating Council (WECC) and
energy resources. supported the Modeling and Validation Working Group.
Spencer Tacke received his BS and MS degrees in electrical He is currently an E&I specialist and Power Engineering
engineering from the University of California at Berkeley, lead at Bayer’s phosphorus facility in Soda Springs Idaho.
c
Mechanical Integrity of Energy Systems Swiss Federal Laboratories for Materials Science and
Technology EMPA Duebendorf,
Switzerland
Abstract Symbols
α Total wedge bevel angle
Overhead line conductors are tensioned to fulfill sag
µ Friction coefficient
limitations. Free-span conditions of common conductors
C1 , C2 Constants for linear fitting function
are investigated and documented in literature. In this
φ Azimuth angle
paper, the structural changes of an all aluminum alloy
A Cross-section
overhead line conductor near a wedge tension clamp
D Conductor diameter
are investigated by x-ray tomography, digital image
d Wire diameter
correlation and finite element analysis. The results
F Tensile force
achieved with these three methods consistently show
l Length
that the inner layers of the conductor are slipping relative
lmech Mechanical transition length
to the outer layers of the conductor. Additionally, not
r Radial coordinate
all wires on the same layer are clamped equally. Thus,
uφ Azimuthal displacement in φ-direction
there is a mechanical transition zone of approximately
ur Radial displacement in r-direction
300 mm between the tension clamp and the free-span.
uz,ref Axial displacement relative to the axial
This mechanical transition zone and the associated
displacement of the reference point uzref.point |10kN
structural changes of the conductor are measured and
of wire 1 on layer 4 at z = 55 mm at 10 kN
modeled. The tools used in this study can be applied on
uz Axial displacement in z-direction
other conductors and clamps as well.
z Axial coordinate
Nomenclature
1. Introduction
Acronyms
1.1. Motivation
AAAC All aluminum alloy conductor
Reliable electric power transmission is essential for the
ACSR Aluminum conductor steel-reinforced
power grid of today and the future. This has evolved
ACT Adapted computed tomography
over many decades and is designed according to local
CCE Clamped conductor end
constraints and preferences. Thus, different kind of
CT Computed tomography
conductors are used.
DIC Digital image correlation
FCE Free-span conductor end On a global scale, aluminum conductor steel reinforced
FEA Finite element analysis (ACSR) are the most commonly used conductors
FIT Fitting curve for overhead lines (OHL). The steel core provides
LW Lower wedge more mechanical strength and is stiffer than the outer
OHL Overhead line aluminum layers. Thus, the conductor is less exposed to
UW Upper wedge elastic and inelastic deformation.
* pascalbuehlmann@outlook.com
KEYWORDS
All Aluminum Alloy Conductor (AAAC), Adapted Computed Tomography (ACT), overhead line, wedge tension
clamp.
A lot of knowledge and literature about the creep of the The conductor temperature in the free-span can be
different kind of conductors with different kind of alloys accurately determined by models summarized in [5].
exist. To additionally benefit from all this expertise, this From the temperature in the free-span, the conductor
paper focuses on methods to determine the mechanical temperature at the clamp can be computed by the model
behavior of individual wires near clamps. The goal of the presented in [6, 7]. With the temperature and tensile force
investigations presented here, is to develop a predictive history or corresponding assumed scenarios, the lifetime
model that allows applying the know-how of creep expectancy of the conductor near the tension clamp can
behavior to specific locations as clamps. be obtained and compared to the situation in the free-
span. It could be quantified, how critical the situation at
At suspension clamps, the conductor is exposed to high the tension clamp can be for conductor creep.
tensile forces on both sides. Thus, the tensile force is
carried by the whole conductor cross-section. But in the Since AAAC are more exposed to creep than ACSR
case of tension clamps, all tensile load is transferred from and other reinforced aluminum conductors, the study is
the conductor to the clamp through the outermost layer performed on a AAAC at a wedge tension clamp, where
of the conductor. In the transition zone between free- all tensile force is transferred from the conductor to the
span with high tensile load and tension free conductor clamp.
behind the tension clamp, the tensile force is transferred However, the here investigated slipping of inner wires,
by friction from layer to layer and through the outermost the conductor bulging due to clamp compression, and the
layer to the clamp. Since there is no form fit, wires of unequally distributed wire stresses due to asymmetrical
the conductor can move relative to each other. Thus, it is clamping may occur on other conductors and clamps as
hypothesized that the inner layers can move relative to well. Especially on wedge type clamps but as well on
the outer layers. This behavior could cause an increased other kinds of tension clamps, these three phenomena
tensile load on the outermost wires which are in contact
may occur, but in varied magnitude. Further testing
with the clamp. Over time, these outermost wires creep
would be required for those applications. On other
and thus elongate more than the ones in the inner layers.
clamp types such as suspension clamps, the investigated
Hence, the inner layers could be loaded more after time
behavior presumably differs significantly. Nevertheless,
which cause these wires to creep as well. This layer
the used techniques can be applied on many other cases.
staged loading is different to the conditions in the free-
In addition, these methods may enable to investigate
span. Because of pitch angle and layer diameter, the
other critical situations of clamped rope structures.
length of wire per unit length of conductor is larger for
the outer layers. Hence, in the free-span the inner layers Especially at suspension clamps, conductors are exposed
are generally loaded more than the outer ones. To enable to fatigue and vibration issues [8–14]. This phenomena
an investigation on this layer staged loading at tension is neither considered nor discussed in this paper.
One x-ray image taken from one angle represent the _____________________
projection of a three-dimensional structure on a two- 1
Golden Engineering, XRS-3
2
Duerr NDT, CR-35
dimensional image. This projection provides insufficient
the image plate is scanned, deleted and re-mounted in pattern is random and unique, which allow automatic
its support on the rotary table, which in the meantime image processing. The speckles and the gaps in between
is rotated by one angle increment. All ten projections are both approximately 3 to 5 pixels large. This pattern
are performed at a constant tensile force. configuration enabled the best pixel subset to analyze
the pattern changes.
• Semi-automatic image processing allows to detect
the steel ball markers and to determine the center of Camera Setup
each marker on each image. The processed images are To observe the spatial deformation of the conductor in
cut into sections along the longitudinal z-axis of the all three dimensions, a stereo camera setup is required.
conductor. The z-coordinate, the projection angle, and Two cameras3 observe the same focus area from two
the distances between x-ray source and image plate different angles. By calibrating the exact position of the
allow to allocate each marker in a spatial cylindrical cameras, the spatial displacement and deformation of the
coordinate system. speckle can be determined.
This procedure is repeated at 10 kN, 20 kN, and 30 kN, To enlarge the focus area, several stereo camera systems
to create three batches of data. The position of the marker were used and the results were merged. To better
center points is changing for each tensile force. This observe the variation of displacement on the conductor
structural change can be represented as displacement of circumference, the stereo camera systems are arranged
each marker as a function of tensile force. at the same longitudinal z-position at different φ-angles.
2.3. Digital Image Correlation To extend the view along the mechanical transition
zone, the stereo camera systems are positioned next to
A second experimental method was required to verify the each other, at the same φ = 90° but with a spacing of
developed ACT method. As strain gauges only measure approximately 55 mm in z-direction.
elongation but no displacement, the optical measurement
with Digital Image Correlation (DIC) was chosen. With Experimental Procedure and Evaluation
this method, speckle patterns on the conductor surface To make the measurement data comparable, the procedure
are compared at varied load conditions. Thus, only the for the DIC measurements is similar to the one with the
outermost conductor layer can be analyzed. However, ACT method. The same conductor and clamp specimen
the discretization is very fine and all surface wires can are used for both cases. First, the speckle pattern was
be observed without any invasive preparation. The DIC applied on the conductor. Second, the cameras were
method and its application in this study is explained in installed and their position was calibrated. The first
more detail in the following. measurement is taken at a tensile force of F = 10 kN.
Speckle Pattern The tensile force was increased to 20 kN and 30 kN.
At both tensile loads another set of data was recorded.
The surface of the conductor is very repetitive and Then, the camera position was changed from lateral to
light reflecting. Thus, it is not suitable to be analyzed circumferential configuration to take another three data
by DIC without any preparation. A white matt primer sets at 10 kN, 20 kN, and 30 kN.
was sprayed on the conductor. Black speckles were ____________________
applied by hand with an airbrush pistol. The resulting 3
Prosilica, GT3400
The model is based on the previously developed model The wedge movement on the α = 5° slope is restricted by
from [1, 2]. This model is validated by single wire and linear equations. To simplify the computation, the upper
entire conductor experiments. wedge and lower wedge are symmetrically displacing on
a slope towards the conductor.
For the investigation done in [1, 2], a model length
of 30 mm was sufficient to describe the mechanical
behavior in the free-span. In the study presented in
this paper, the mechanical transition zone between the
wedge tension clamp and the free-span is investigated.
Therefore, a much longer FEA model is required.
The complete length of clamped conductor
lclamp = 275 mm plus additional conductor of 700 mm
including the transition zone are required. This results The grooved wedge surfaces cause a form fit between
in a model length of 1000 mm. This model is no longer wedges and conductor. Therefore, it is assumed that
solvable with an implicit solver as done previously in established contacts between wedge surfaces and wire
[1, 2]. Thus, an explicit quasi-static algorithm is chosen. surfaces cannot be released anymore.
The explicit method was verified by comparing the The clamped conductor end (CCE) on the non-tensile
explicit and implicit approach on a load case solved by load side protrudes the wedges by approximately 25 mm
both methods. As for the ACT and DIC experiments, when clamped (see Figure 6). To prevent free-body
the axial displacements uz serve as comparison to the motion, all CCE nodes are constrained in r, φ, z-direction
explicit solver approach. The difference in displacement during Step 1 of the analysis. From Step 2 on, all CCE
uz between the explicit and implicit approach is in a nodes are released in the longitudinal z-direction:
range of 3 %. This deviation is acceptable because the
implicit solver approach was confirmed in [1] to have
force-elongation characteristics in a range of less than
5 % difference compared to analytic models as well as
experimental measurements.
The displacement of all nodes on the free-span conductor
As shown in Figure 5, all 61 wires are discretized end (FCE) are restricted in circumferential φ-direction.
* sudhakar.Cherukupalli@bchydro.com
KEYWORDS
HDD, Cables in Ducts, Horizontal Directional Drilling, Seismic Performance of Cables, Dynamic Analysis
(b)
Figure 1a) – Plan and profile of a typical HDD under a body of water b) Cross-section of this horizontal direction drill
the cable inside the duct and manholes at each end are de- with the guide posts and entry/exit ducts is modelled by
coupled until the pre-determined slack of 100 mm or 1.0 m assigning contact elements. The straight joint and cable
is reached. Once the effect of the cable slack configuration duct-to-manhole interface are modelled so that interface
is investigated, this can be inserted as a feed-in to the local loads can be calculated.
manhole model as shown in Figure 3. The cable interaction
Fig. 2: Assumed geometric profile at the North and South ends of the HDD crossing
A. Seismic Data
It is estimated that the velocity of the seismic wave
propagation for compact to dense sand/silt stratum and
weak sandstone bedrock is 1,500 m/s. The arrival times
for the seismic wave along the cable are defined in Table 2
and compared to a wave propagation velocity of 500 m/s.
Table 2: Arrival time of seismic wave along cable for different wave
propagation
120 0.24 0.08 Typically, cables can experience high tensions during
installation, which reduces with thermo-mechanical
240 0.48 0.16 relaxation forces during cyclic loading within the duct.
360 0.72 0.24 Therefore, as part of the dynamic analysis, the expected
initial conditions of the cable have to be determined. This
480 0.96 0.32 can be done by pulling the cable through the ducts and,
subsequently, artificially expanding and contracting the
600 1.20 0.40
cable to simulate the relaxation in the cables once put in
720 1.44 0.48 operation. Previous studies did not investigate different
initial residual tensions of the cable system prior to arrival
For this study five earthquake time histories were evaluated. of seismic wave. Thermo-mechanical loading was also not
The time histories were selected based on the obtained considered in the original study.
Uniform Hazard Spectra (UHS) curve for 10% damping.
B. Cable Modeling Methodology
These were scaled to match the spectral acceleration in
the range of estimated fundamental periods of the cable The 3D model in this study consists of 47 contact
model of 0.15 to 0.2 s. An example of one of the applied elements to simulate the seismic motions of the duct
earthquake time histories is shown in Figure 4. system. Each duct section has an imposed displacement
that takes into account the arrival time of the seismic by the axial spring-damper at the center of each segment,
wave. The cable is placed inside the duct sections and which applies an equal and opposite effective tension
consists of 1720 line elements. Springs are attached at force to the nodes at each end of the segment. The bending
each end to model the cable slack and interaction with properties are represented by rotational spring-dampers on
the manholes, see Figure 5. either side of the node, spanning between the node’s axial
direction Nz and the segment’s axial direction Sz.
The cable is modelled as line elements as shown in Figure
6. Each line consists of a straight massless segment with C. Dynamic Analysis
axial and torsional properties. The mass, forces and
The dynamic response of the cables is determined by
moments are placed at the nodes.
solving the equation of motion in the time domain
The axial stiffness and damping of the line are modelled according to the following expression:
In general, Figure 8 shows that the peak tensions reported Additional sensitivity analyses show that the distribution
along the cable are between -20 kN and 20 kN for the of the tension is also affected by the direction of seismic
100 mm slack load cases. Furthermore, it is shown that wave propagation. In general, higher tensions are found
there is not a significant difference between the peak at the downstream end of the seismic wave. During detail
tensions reported for the two slope configurations. On design of the cable system it is therefore important to also
average, the maximum reported tensions along the cable include directional effects – either by expanding the load
lengths increase by 5% when the angles increase by 5°. case matrix or running several sensitivity cases that take
The difference between the peak compression loads is into account the potential approach angles of the seismic
typically less than 1%. A similar comparison for load cases waves
with 1.0 m cable slack at manholes showed similar results, To investigate the sensitivity of the dynamic response of the
albeit with peak forces computed along the cable typically cables to seismic wave propagation velocity, the 1,500 m/s
in the -20 kN to 10 kN range. is compared to a lower bound value of 500 m/s. Using the
Figure 9: Comparisons of tensile and compressive forces for the two different seismic propagation velocities
model developed for the 18°/15° slope configuration, approach slopes and considerable savings on materials and
the results are summarized in Figure 9. construction activities and possible improved construction
schedule. Discussion will need to be held with the drilling
From this sensitivity check it is shown that the arrival time
Contractor to confirm that such a drill profile can actually
(hence the seismic wave velocity) of the seismic wave
be achieved practically. Simultaneous consultations with
has a more significant impact on the peak tension results
a qualified Geotech engineer have also to be considered
compared to increasing the slope angles of 10° and 13°
before a final decision can be made.
by 5 degrees.
By comparing the results from the 10°/ 13° and 15°/ 18°
From Figure 10 it is shown that the cable axial movement slope angles, it is shown that there is not a significant
at the cable end follows the ground motion closely. The difference between the peak tensions in this 45 m deep
graph illustrates how the cable slack of 100 mm is reached crossing. On average, the maximum tensions along the
after approximately 4 seconds from arrival of the seismic cable lengths increase by 5% when the angles increase
wave and, after some initial slippage, follows the general by 5° under a coefficient of friction range of 0.18 to 0.23.
ground motion. When the installation is new the expectation is that µ; the
The original analysis was based on a horizontal axial seismic coefficient of friction will be low but with the passage of
excitation only. In order to assess the full dynamic response time, introduction of sediments into the pipe will tend to
of the cable system, vertical and lateral components need increase the value for µ. The values selected were based on
to be added to the time series. By including combined a separate study conducted to measure static and dynamic
axial and vertical motions, where the vertical component cable friction in pipes with and without water.
represents 80% of the horizontal motion, the peak forces From a sensitivity study of the seismic wave propagation
are found to increase by approximately 25% and 18% velocity, it is shown that the arrival time of the seismic
under tension and compression, respectively. wave along the cable length can have a more significant
impact on the peak tension results compared to the 5°
Conclusions and recommendations change in slope angle.
By performing numerical modelling of the seismic By subjecting the cables to thermal expansion and
response of transmission cables and accessories, the contraction prior to running the seismic events, it is shown
anticipated tension and compressive forces experienced that the load cases that represent the lower bound values of
in the cable and joint can be better estimated. This may friction and axial stiffness are highly sensitive to the initial
be used to appropriately tailor and optimize the seismic residual tension conditions.
testing protocol and qualification of a cable system.
The original seismic study only considered axial seismic
Based on the improvements made to the seismic response ground motions. By including a vertical component
models, it is shown that the approach angles at each end that represents 80% of the horizontal motion, the peak
could be increased by up to 5 degrees. For an 850 m maximum and minimum forces change by +25% and
crossing this would result in a horizontal shortening of the -18%, respectively.
* jiajia.liu@strath.ac.uk
KEYWORDS
Denoising, detection, partial discharge, wavelet selection, wavelet entropy
h
Scale 1 2 the wavelet domain [9], [16], [17]. General procedures
h Approx. cfs
2
Low-pass filter
Downsampling
for the wavelet-based denoising of a signal are presented
as follows:
g
Input data Low-pass filter
Downsampling 2
g
2
High-pass filter 1) Apply DWT to decompose the noisy signal s with a
Detail cfs
High-pass filter selected wavelet to a predefined scale J, and obtain
Figure 1. The implementation of DWT in signal decomposition approximation coefficients aJ at the final scale J and
detail coefficients dj at decomposition scale j, where
Scale 2
~
h
j = 1, 2, ..., J.
2 Scale 1
2) Estimate the threshold through a noise estimation
Approx. cfs ~
h
2 technique and apply this threshold to detail
Upsampling Low-pass filter
~ Output data
coefficients, dj, at decomposition scale j using hard
g Low-pass filter
2 Upsampling
~
High-pass filter
Detail cfs
2
g
or soft thresholding scheme.
High-pass filter 3) Apply IDWT to the approximation coefficients aJ
Figure 2. The implementation of IDWT in signal reconstruction and the processed detail coefficients d'j to reconstruct
the denoised signal s'.
Equally, the implementation of DWT in signal
Based on the noise estimation technique proposed in [9],
decomposition also can be interpreted in the frequency
[18], the scale-dependent threshold used in this paper is
domain. As shown in Figure 1, the DWT process is
estimated by
equivalent to filtering the signal by the filter pairs, the
low-pass filter h and the high-pass filter g. Ideally, these (2)
filter pairs halve the frequency band with the increase
of scale. Let fs be the sampling frequency of the input
where MAD|·| is the median absolute deviation of the
signal, the frequency band, G1(w), of the output of the
detail coefficients dj at decomposition scale j, and nj
high-pass filter is fs/4 ~ fs/2, while the frequency band,
is the length of dj . For the thresholding scheme, soft
H1(w), of the output of the low-pass filter is 0 ~ fs/4.
thresholding in [18] is used in this paper, the function
For next scale, H1(w) is further split into G2(w) and
is given by
H2(w), which are fs/8 ~ fs/4 and 0 ~ fs/8, respectively. The
frequency band is iteratively halved in the subsequent
(3)
decomposition in the same manner until the predefined
scale reaches. For a J-scale DWT, the distribution of the
corresponding frequency bands is shown in Figure 3. It where i = 1,2, ..., nj.
(5)
(6)
ers used in (8) and (9) [10].
Similar to Shannon entropy, wavelet entropy is applied
to measure the degree of disorder of wavelet coefficients
The simulated sampling frequency fs is set to 60MHz.
or signify the randomness of wavelet coefficients. It is
Figure 4 shows these two simulated PD signals both in
important to note that wavelet entropy is not an information
time and frequency domains. Generally, DOP signal
cost function (ICF), since it requires the energy of wavelet
shown in Figure 4 (c) and (d) is closer to a real high-
coefficients to be normalized as shown in (5), and is thus
frequency PD signal detected from electrical power
not additive [13], [14]. Substituting (5) into (6) yields:
equipment in practice [10]
To develop the new scheme for practical use, real PD
signals were generated through an artificial defect of a
7mm × 7mm breach in the outer conductor created in a
1.5m 11 kV ethylene propylene rubber-insulated (EPR)
cable sample [20]. PD signals were collected using a high
frequency current transformer (HFCT). The specifications
of the HFCT are listed in Table 2. Details regarding the
experiment setup are depicted in Figure 5 [20].
Table 2. Specifications of the HFCT.
Parameters HFCT
Sensitivity 5 V/A
-3 dB bandwidth 90 kHz – 20 MHz
(7) Internal diameter 50 mm
External diameter 110 mm
In (7), l is an ICF based on the definition in [13]. As such,
Load resistance 50 Ω
wavelet entropy is a monotonic-increasing function of
l, which means minimizing l over wavelet coefficients Output conductor BNC
minimizes wavelet entropy. Manufacturer IPEC
Stress Artificial
Voltage cone defect
Zn probe
Cable sample
Ck Earth Strap
HFCT
Tr
Figure 5. PD testing of a defective 11 kV EPR cable. HFCT was used to collect PD pulses (Ck and Zm
represent the coupling capacitor and measuring impedance respectively).
Experiments were performed at various voltage levels. in this paper. One PD pulse, named , with 2048 sample
The PD pulses measured at 9kV are used as the real PD points was selected and depicted in Figure 6.
signals to demonstrate the new wavelet selection scheme
Figure 6. Real PD pulse, s3, detected from a defective EPR cable under 9kV AC voltage
(11)
(10)
where i = 1, 2, ..., ni, ni , is the length of approximation
where is the mean of xi and is the mean of yi. The coefficients or detail coefficients at scale j, and dj is the
value of γ is in the range of -1 to 1. It takes on a valueclose detail coefficients at scale j.
to 1 indicating and are positively correlated, and a value
The idea of wavelet energy was introduced in EBWSS.
close to -1 denoting they are negatively correlated. A
For an orthogonal wavelet, energy preservation is one of
value of γ near zero means xi and yi are uncorrelated.
the desirable properties of DWT [21]. The equation for
For CBWSS [9], correlation is used as a measure of the energy preservation is given by
similarity between a pure PD signal and a wavelet, and
this similarity is referred to as their shapes. The more (12)
similar their shapes, the higher the correlation coefficient
is. The wavelet that has the highest correlation coefficient where α and d are the approximation and detail
with the shape of a PD signal is selected to maximize the coefficients of the DWT of a signal X. This property is
wavelet coefficients through wavelet analysis. also applied to PD signals using DWT decomposition. A
PD signal can be decomposed into J scales with signals,
The general process for the choice of appropriate wavelet
i.e., s1, s2, ..., sJ, sJ+1. Among these signals, s1, s2, ..., sJ, are
using CBWSS is described as follows:
detail coefficients from scale 1 to scale J, while sJ+1 is
a. Analyze the detected PD signal to generate a ‘typical’ the approximation coefficients at scale J The energy of a
PD pulse. decomposed signal sk is given by
b. Set up a wavelet library, consisting of the wavelets that
have similar characteristics to the PD pulse.
(13)
c. Normalize the PD pulse and each wavelet retrieved
from the wavelet library. where k = 1, 2, ..., J + 1, i = 1, 2, ..., ni, and ni is the length
d. Calculate the correlation coefficient, , between the PD of si. Then, s can be represented by a normalized energy
pulse and each wavelet. vector (e1 , e2 ,..., eJ, eJ+1), where ek is defined as
e. Select the wavelet that has the maximum correlation
coefficient with the PD pulse, it will be applied for the (14)
following wavelet-based denoising.
As indicated in Section 1 the CBWSS approach is limited It can be seen that the concept of energy ratio in EBWSS
by noise and is scale-independent. Also, a heuristic method can be interpreted as a normalized energy vector. Figure 7
was introduced in [10] to obtain better correlation results. shows the DEP, DOP, and white Gaussian noise (WGN)
Resampling both the PD signal and wavelet function in used in [10] to explain the criterion of EBWSS for wavelet
time domain is applied to align their peaks as well as their selection. Figure 7 (a) and (b) show the DEP signal and
first zero-crossing points after the peaks. This method is its normalized energy vector respectively. Equally, Figure
7 (c) and (d) show the DOP signal and its normalized decomposition, and select the optimal wavelet based
energy vector. Figure 7 (e) and (f) show WGN and its on the strategy used in steps a, b, and c.
normalized energy vector. Based on Figure 7 (b), (d) and e. Iterate the steps above until the predefined
(e), the approximations of the DEP and DOP signals cover decomposition scale J reaches. The optimal wavelet
the most energy of total coefficients while the details of for each decomposition scale will be selected.
WGN preserve the most energy of total coefficients [10]. Results in [10] show that EBWSS outperforms CBWSS
The general process for the choice of an appropriate for the best wavelet selection. However, EBWSS is not
wavelet using EBWSS is presented as follows: as robust as expected. It selects the wavelet that can
a. Given a wavelet library {ψi:i= 1, 2, ..., N}, select a maximize the energy ratio of approximation coefficients.
wavelet from {ψi}, and perform a one-scale DWT It is not strictly true for DOP signals, particularly when
decomposition of a noisy PD signal. Obtain its the decomposition scale increases. It can be seen from
approximation coefficients a1(i) and detail coefficients the normalized vector of DOP in Figure 7 (d), the energy
d1(i). of PD signal with a 6-scale decomposition is preserved
b. Calculate the energy ratio of approximation coefficients on the detail coefficients rather than approximation
coefficients. When more scales are required, for example,
based on (11). If is the maximum of
7 scales, the EBWSS is still trying to select the appropriate
, ψp, select as the optimal wavelet wavelet by maximizing the energy ratio of approximation
for the first scale. coefficients.
c. Apply ψp to obtain the approximation coefficients a1(p) The limitation of EBWSS can be interpreted in the
and d1(p). frequency domain. Based on Parseval’s theorem, the time
d. a1(p) is used as the input signal for next-scale DWT and frequency domains are equivalent representations
the performance of a proposed denoising method or and then, various wavelet selection schemes are used
algorithm. ME, MSE, and XCORR are used in this to remove the noise and evaluated by the parameters
paper to compare the denoising results of different mentioned above. As mentioned in Section 4.2,
wavelet selection schemes. XCORR is calculated based EBWSS is not strictly true when the decomposition
on (11), and ME, MSE are calculated by the equations scale increases over 6. To highlight the limitations of
as follows, EBWSS, the decomposition scale is set to 7 for the
results analysis. Two simulated PD signals, s1 and s2,
and real PD signal, s3, as well as their noisy signals NS1,
(15) NS2 and NS3 with SNR = -10 are depicted in Figure 10.
It is important to note that the original real PD signal
(16) shown in Figure 6 is corrupted by ambient noise during
experiment. To mitigate the effect of this noise on the
where m and m' are the magnitudes of s(i) and s'(i) denoising results, it has been pre-processed using the
respectively. s(i) represents the original signal and s'(i) method introduced in [10]. The smoothed real PD
denotes the denoised signal. N is the length of signals. signal is depicted in Figure 10 (e).
Better denoised results can be obtained with lower ME,
Denoised S1, S2, and S3 using CBWSS, EBWSS, and
MSE, and higher XCORR.
WEBWSS are shown in Figure 11 (a), (b), and (c)
In this paper, PD signals are corrupted by white noise, respectively. The related parameters used to evaluate
Figure 11. (a): Denoised s1 using CBWSS, EBWSS and WEBWSS, (b): Denoised s2 using
CBWSS, EBWSS and WEBWSS, (c): Denoised s3 using CBWSS, EBWSS and WEBWSS.
Figure 13. ME, MSE and XCORR between s2 and denoised s2.
Figure 14. ME, MSE and XCORR between s3 and denoised s3.
Figure 15. The denoising results of on-site PD signal: (a) On-site detected PD signal; (b), (c), (d) are the denoised PD
signal by CBWSS, EBWSS, and WEBWSS, respectively.
H. KONARA*, U. D. ANNAKKAGE**
University of Manitoba, Canada
C. KARAWITA***
TransGrid Solutions Inc., Canada
This paper presents a co-simulation model to enable would require a lot of parallel processing hardware which
simulation of large power systems in real-time. One is expensive. On the other hand, Transient Stability (TS)
approach for co-simulation is to model a small part of the models use a larger time-step (half a cycle) and simpler
network in detail using an EMT simulation and model the models compared to EMT simulations delivering faster
rest of the network using a Transient Stability (TS) model. simulations. As a result, TS models are used to simulate
To capture the high-frequency behavior of the system, very large power systems. Because of the larger time-step
Frequency Dependent Network Equivalents (FDNE) are and simpler models used, TS programs are only accurate
used at interfacing busses. This paper proposes a different enough to model slow electromechanical transients. The
approach where the TS model is replaced with a Dynamic need of a simulation model that is capable of simulating
Phasor (DP) model. Since the DP model covers a wider a large power system while accurately capturing high
bandwidth, an FDNE is not needed. The numerical frequencies is overcome by co-simulation models (also
instability of a real-time EMT-DP interface due to time- known as hybrid simulations) [1].
step delay is overcome by using a data extrapolation Interfacing transient stability-type simulation models to
technique. The proposed co-simulation platform was EMT simulation for analyzing large power systems has
tested in real-time with detailed synchronous machine been reported in [1]-[4]. Typically, a small part of the
models in both the EMT and DP parts of the network. network containing power electronic devices is modeled
in detail at the switching level using an EMT model
1. Introduction and the rest of the system (external system) is modeled
The Electromagnetic Transient (EMT) simulation is using a TS model. Since EMT is a high-frequency model
the most accurate tool available for simulating power and TS is a low-frequency model, interfacing them into
systems. EMT simulations can capture dynamics in a one simulation platform is not possible without using
broad frequency bandwidth and therefore, can accurately Frequency Dependent Network Equivalents (FDNE) at the
model fast dynamics such as switching, dynamics of boundary buses [5]. In this approach, TSA represents the
HVDC systems, etc. However, EMT simulations can be slow electromechanical behavior of the external network
computationally demanding since it uses a small time- and FDNE represents the high-frequency behavior of the
step (50μs or less). Due to the computational burden of external system. FDNE has a fitted admittance. A rational
EMT models, a simulation can run for an unaccceptable function is derived from the frequency response of the
duration making EMT time simulation unsuitable for a system using curve-fitting techniques. A lumped parameter
large power system. Real-time simulation tools use EMT model is then derived from the rational transfer function
modeling and updates its values each time-step in real using a recursive convolution method. FDNE improves
time; i.e. 50μs in the simulation is 50μs in the real world. the accuracy of the EMT-TS simulations by representing
Using real time simulation for a large power system the high-frequency behavior of the TS network. However,
*konarhkk@myumanitoba.ca
**udaya@ieee.org
***ckarawita@tgs.biz
KEYWORDS
Co-simulation, dynamic phasors, electromagnetic transient, real-time, transient stability.
Hence, dynamic phasors add the derivative of the phasor higher than the stator impedance and therefore Norton
signal to the derivative modeled in transient stability equivalent-type synchronous machine presented in [30] is
studies. Using this property, differential equations for a not stable. Hence, in this work, Thevenin equivalent type
capacitor current and an inductor voltage can be written [32]-[36] synchronous machine model is used to represent
as (3) and (4). the machines in the DP side of the network.
(5)
(6)
(7)
(8)
* shrutidwarkanath.rao@ge.com
KEYWORDS
Damping ratio, Eigensystem Realization Algorithm (ERA), electromechanical oscillations, modal identification, Time-
domain simulation
could be incorrectly flagged as being poorly damped. Such that the modes identified for these two signals are very
signals where FFT results, singular values etc. are not clear accurate.
indicators of the true modes of the system, often do not
Spurious modes do not contribute much to the signal
exhibit a clear oscillatory behaviour. This will be discussed
shape. An example of this is shown in Figure 3, where the
in detail in latter sections of the paper.
plot on the left has the signal reconstructed from a system
A good measure of the validity of the identified modes with seven eigenvalues while the signal on the right has the
is whether a signal constructed using the identified same signal reconstructed from five eigenvalues. It can be
modes approximates the signals from which the modes seen that the spurious mode i.e. the last pair of eigenvalues
are derived. Examples of such comparison are shown does not contribute much to the shape. In some rare cases,
in Figure 5, where the blue curves represent the impulse the inclusion of a spurious mode can make the match
response of the linear system while the black curves are between the actual signal and reconstructed signal worse.
the true signals used for modal identification. Based on This is shown in Figure 4 where the plot on the left has a
the excellent match between the two, it can be concluded match with 13 modes, while the plot on the right has the
Figure 5 Examples of cases where the signals constructed using the modes match the true signal accurately
WECC Disturbance Time- Total number A: Number of B: Number of Comments regarding poorly identified
case frame of of signals signals where signals where cases (both categories A and B)
study studied spurious mode chosen number
was selected modes are
insufficient for
accurate match
Case1 3ph bus fault 8-10s 220 0 1 Spread less than 0.1 MVAr
Case1 Step voltage 8-10s 220 2 2 Very flat/small
reference of a large
nuclear generator
Case1 3ph bus fault 5-10s 66 7 2 Very flat/small
Case2 Trip large 6-15s 66 13 11 11/13 signals with spurious mode:
generator signal very small. 8/11 with poor match
are signals that are very small
Case2 Trip large generator 10-15s 66 8 0 Very flat/small
Case2 Trip load 6-10s 66 9 5 All too “chattery”/nonlinear with low-
magnitude variations
Case2 3ph linear fault 6-10s 66 5 0 2 signals have significance metric close
to threshold, all signals less than 10
MW/MVArs
Case2 LLG bus fault 6-10s 66 2 0
Case2 LG line fault 6-10s 66 1 1
Case2 3ph bus fault 10-13s 66 1 0 Signal almost linear, no oscillations
Figure 7 Example where higher number of modes are needed for accurate modal identification
Other problematic situation occurs when the variations are approach, the automated modal identification tool
so small that the signal appears to be noisy, for example described in the paper, IDtools has been integrated in a
see Figure 9 (The left image shows the signal over the transient stability simulation program. For each signal
entire simulation duration while the zoomed in image being processed, it identifies up to three oscillatory
focusing on the last 2 seconds of simulation is shown on modes, out of a total of 7 eigenvalues (including purely
the right, variations in the last 2 seconds are of the order of real eigenvalues as well as complex-conjugate pairs). If
0.0001 MVAr). The program should not be used to process appropriate signals with sufficient signal variations were
signals that are noisy and/or exhibit “chatter”, as the used, it appropriately identified approximately 98% of
signal constructed from the modes will not match the true nearly 900 signals inspected. With a C implementation, it
signal well due to the presence of sudden variations in the takes on average 0.12 s to process each signal, with a major
signal (which are physically meaningless). Such signals portion of the time spent in reading and pre-processing the
can be detected by metrics such as signal to noise ratio. data. Since all signals are processed independent of each
The program should not be used for signals that exhibit other, the process can easily be parallelized for speed.
discrete and/or abrupt changes such as those encircled in This is the first step towards a complete automation of
Figure 10, since the signal constructed from the identified modal identification from time-domain signals. These
modes cannot capture the abrupt changes in the signal. computations are strongly recommended to be done after
This is an indication that the identified modes might be all automatic discrete adjustments in the system have
in error and the RMS errors between the actual signal and already been made following a fault clearance or other
the reconstructed signal will consequently be higher. major switching event, in order to ensure that the signals
do not have any sharp changes. The process of filtering
6. Conclusions out unnecessary modes is a key step which will continue
This work is the latest effort toward merging modal to be refined. While the screening is not 100% accurate,
damping identification from time domain simulations users can look at the predominant modes for several
in a commercial product. Using the described screening signals in a given area since they are expected to have the
*Nayeem.MohammadAbdullah@ComEd.com
KEYWORDS
Distributed energy resources (DERs), locational marginal value (LMV), programming tools, power system
simulations
generation but also via local or community microgrids to be addressed. Generally, for radial distribution circuits
that provide, in effect, a form of backup generation to the DER must be downstream of the location of a loading
one or multiple customers. DERs can provide these issue to help mitigate the problem. Figure 1 illustrates
benefits via the supply of real power, reactive power, and the DER overload mitigation concept through a simple
reserve capacity. Once DERs become a significant part of example. The DER downstream of the congested line
the distribution system portfolio, some DERs may also provides more benefits in comparison to the DER
provide a reliability benefit in terms of local reserves; that upstream of the congestion. Similarly, voltage problems
is, additional real and reactive power capacity available to on circuits are typically local in nature. Low voltage at
provide capacity, voltage, or reliability in case the DERs the end of the circuit does not imply low voltage near or
primarily providing that service become available. at the station. Indeed, attempting to cure a low voltage at
DERs pose opportunities as well as challenges. one location by raising overall circuit voltages may cause
Unmanaged DERs are often a source of power quality too high a voltage in another location on the same circuit.
problems, especially high voltage and flicker if power Traditional mitigation to voltage issues are to deploy local
electronically-coupled, and possibly circuit ampere apparatus that affects voltage locally, such as capacitors or
overloads in cases of extreme reverse flow [2], [3]. DERs voltage regulators.
can, however, be a means to mitigate these problems, even If we examine the simple case of a radial feeder shown
increasing hosting capacity for other DERs, including EV above where the first mile of a circuit needs to be reinforced
charging by providing local real and reactive power to and the argument that DER only affects loadings on the
mitigate the adverse effects of locally high amounts of PV upstream portions of the circuit, then this introduces the
or other variable DG. concept of how to place a locational value on the DER. If
We first discuss the fundamental aspects of the value of the DER midway on the overloaded portion of the circuit
DER and the nature of the problem focusing on distribution deferred the need to upgrade capacity on the first half
avoided cost. The key contribution of this work is to of the circuit – conceivably it could at maximum avoid
provide a generic algorithm for direct calculation of the half of the cost. The DER located at the downstream end
value of DER that is technology agnostic and is based of the affected portion of the circuit could, by contrast,
on the AC power flow formulation. We present the avoid all the cost. So conceptually from a distribution grid
preliminary results of two test systems. perspective, the second DER offers twice the value of the
first. This concept can be generalized. For any DER, its
2. Value of DER Fundamentals locational value is affected by the sensitivity of the circuit
condition requiring capacity or voltage investments to the
The first principle in determining the value of DER to the
real and reactive power from that DER.
distribution grid in terms of avoided costs is that all DER
value is locational [4]. That is because all avoided costs Another concept is important when we evaluate the value
for capacity and voltage are investment costs in response to the distribution grid of specific DER technologies.
to actual forecasted needs of the system, which are always Circuit capacity and voltage problems typically do not
based on specific grid issues in specific locations. On a happen to the same extent across every hour of the year,
given circuit, the ability of a DER to avoid capacity or and in fact may only occur for a limited number of hours
voltage costs by providing real or reactive power depends a year. An effective and efficient DER technology would
very much upon the location relative to the grid constraint essentially provide real and reactive power when and
step in the process is to implement a “reduction” in the on branch 1-2.As it can be observed from the reported
specified distribution network considered in the study. results, the LMV for real power is much higher than that
The reason for this pre-step is that the proposed method for reactive power. This is due to the load power factor.
is designed for balanced, single-phase networks and as The branch overload can be relieved when real and
such, elimination of two-phase or three-phase elements reactive power is injected downstream the location of the
is required for the appropriate power flow solution. The violation, as shown in Figure 1. Moreover, the LMV for
“network reduction” step also aims to simplify the network both real and reactive power seems to be increasing as
and remove redundant elements and loops to expedite the the electric distance from the violation increases. This
computation procedure. Furthermore, the removal of very is intuitively correct, as DER further away would also
small element impedances is required to avoid numerical reduce additional losses. In Table 2, the simulation results
issues in the computation of the admittance matrix and for modified current flow limits such that only branch
the Jacobian elements. The “network reduction” step is 2-3 experiences violation have been presented. A few
particularly useful when dealing with large-scale realistic interesting observations can be made here. First, the LMV
distribution feeder topologies. for real and reactive power has dropped significantly
Step 2 of the process is to specify the real and reactive at node 2 because the current limit for branch 1-2 has
nodal demand in the network to solve the AC power flow. more than doubled compared to Table 1; and the amount
This is accomplished by first computing the Jacobian of violation has dropped to zero, implying that the per
matrix around an initial state of the system. The Newton- unit MCC value has decreased accordingly. The LMV
Raphson method is implemented for the AC power flow is not exactly zero because it accounts for the losses in
solution at each iteration and the desired solution is derived the network. As expected, the LMV for real and reactive
when the mismatch error is below a certain threshold. The power are significantly higher for DER located at node 3
solution of the power flow are the state variable, i.e., the where it can contribute to alleviate the overload.
nodal voltage magnitudes and angles are used. In Step 5, The second set of results is based on the analysis of
the power flow solution is compared to current and voltage an IEEE 33-bus network. This case is set up such that
limits to identify the location, number and the magnitude of thermal overloads are exist on branches 1-5, as shown in
violations in the system (voltage violations or overloads) is Figure 4. We also set the MCCT to be $12,000 annually
determined. Since the MCC can be readily obtained from to represent the annualized cost of upgrades. The resulting
the power flow solution as indicated by equation (2), the real and reactive LMVs are as shown in Figure 5 for the
LMV for real and reactive power can be computed in Step peak hour. It is observed that both real and reactive LMV
6, based on equation (3). The process can be repeated for increase as we move down the circuit down the circuit
each hour in the planning horizon. such that DER at the nodes will affect more and more
overloaded branches. For all the nodes downstream of the
5. Test Cases Results overloads, the LMV increase very slightly and only due
In this section, results from two networks are presented. to losses. Lower LMVs in nodes 19-25 are due to limited
The first set is from a simple three-bus network shown contribution of these nodes to relieving congestion in
in Figure 3. The power flow results and the branch flow upstream branches 1 or 2. It is also important to point that
overloads are provided in Tables 1 and 2. The tolerance reactive LMV shows the value of reactive injections at
for the solution is equal to 1E-5. In Table 1 we impose each node in relieving overloads by improving the power
branch current limits such that there is an overload only factor.
Table II. Direct LMV calculation results for overload on branch 2-3.
2
Central Research Institute of Electric Power Industry
Japan
Summary and wind power generation plants, are one of the serious
threats to the safety and stable operation of them. Hence,
Lightning strikes to electric power facilities with tall
effective and reasonable countermeasures against the
structures often disturb the safety and stable operation of
lightning strikes to each facility are required. In addition,
these systems. Immediate identification of the lightning
it is necessary to identify the lightning struck facility
struck facility is important to perform speedy and appropriate
immediately to perform appropriate recovering works.
recovering works. Accurate locating of strike points
Lightning location systems (LLSs) are operated by, for
provided by lightning location systems (LLSs) contributes example, electric utility companies, and they have become
to preparing speedy and suitable countermeasures against indispensable tools for increasing the reliability of the
those accidents. power system. Accurate locating of strike points provided
Lightning electromagnetic waves are delayed and deformed by LLSs contributes to preparing speedy and suitable
by the earth surface conditions such as soil resistivity. It is countermeasures against lightning accidents.
desirable to correct the propagation delay due to the earth The delay and deformation of the lightning electromagnetic
surface conditions for improving the accuracy of the LLS. waves are affected by the earth surface conditions such as
Effects of ground surface conditions, especially the soil resistivity. The propagation delay should be corrected
horizontally stratified ground case, on propagation according to the conditions [1], [2]. In order to correct the
characteristic of a lightning electromagnetic pulse (LEMP) propagation appropriately, the effect of ground surface
are analyzed by the finite-difference time-domain (FDTD) conditions on propagation characteristic of lightning
method in this paper. In order to simulate the ground electromagnetic pulse (LEMP) waveforms need to be
condition in a rainy day, soil resistivity within 1 m below the clarified.
ground surface is set a value lower than that of the second This paper investigates the effect of ground surface
layer. Since the 1 m thick ground surface layer with soil conditions, especially horizontally stratified flat ground [3].
resistivity lower than some tens of Ωm acts as a shielding The effect of the stratified ground, whose soil resistivity of
layer, LEMPs cannot penetrate into the second layer. Thus the first layer is higher than that of the second layer, has
the propagation speed is little reduced by the surface with been analyzed (e.g. in [3]) since that effect enlarges the
low resistivity. The over-correction for propagation delay magnitude of the induced voltages on distribution lines
based only on a high soil resistivity assuming a sunny day located at close or intermediate distance from the lightning
can result in the decrease of the location accuracy since strike point. The characteristic of the opposite condition
precipitation reduces the soil resistivity of the surface layer. (with a lower resistivity in the first layer), however, has
not yet been clarified. The surface layer can have lower
1. Introduction resistivity in rainy days, and this surface condition might
Lightning strikes to electric power facilities with tall decrease the accuracy of propagation delay correction
structures, such as transmission and distribution systems based on a conventional method. These characteristics are
* cyjd1302@mail4.doshisha.ac.jp
KEYWORDS
Lightning, LEMP propagation, stratified ground, lightning location system, FDTD method
(5)
where, ε0, c0, and vch are vacuum permittivity, light speed,
and lightning-current propagation speed along a lightning (6)
channel, respectively.
The first, second, and third terms in the above equation
are electro static, induction, and radiation components,
respectively. The LEMP waveform on the flat ground where ρuni and εruni are resistivity and relative permittivity of
beyond some tens of kilometers, which is used for LLS, the uniform ground.
can be regarded as the radiation component, and is written For the horizontally stratified ground case, the normalized
in frequency domain as follows. surface impedance Δstr and associated coefficients are given
as follows [3], [7].
(2)
(7)
where ω is angular frequency and j is imaginary unit
(8)
where I01 = 11 kA, I02 = 7.5 kA, η = 0.717, τ1 = 0.75 μs, τ2
s, τ3 = 100 μs, and τ4 = 6.0 μs, respectively.
= 5.0 μ
The 7-km high straight vertical lightning channel is
expressed with the lossless transmission line model.
The propagation speed of the lightning current along the
lightning channel is set to the half of the light speed. Vertical
Figure 1. Geometry used in deriving approximate theoretical formula.
electric field waveforms at horizontal distance of 20 and 44
km from the lightning struck point are calculated.
2.2 FDTD Model
Figure 2 shows the FDTD model in the 2-D cylindrical
coordinate system for calculating LEMP propagation over
the uniform or horizontally stratified ground [8]. An r-z
plane of the FDTD calculation space is divided uniformly
into 1×1 m square cells. The soil resistivity of ground is
set from 0 Ωm to 1 kΩm in the uniform ground case. In
the horizontally stratified ground case, soil resistivity
of ground within 1 m below the surface (1 cell) is set to
a value ranging from 1 Ωm to 500 Ωm, and that for the
second layer is set to 1 kΩm. The relative permittivity of Figure 2. FDTD analysis model of LEMP propagation over horizontally
stratified ground. FDTD analysis space is divided into two regions to employ a
soil is set to 10 in all cases. moving-window technique.
(a) (b)
Figure 3. (a) Vertical electric field waveforms and (b) their time derivatives calculated with the FDTD method and theoretical expression in the
case of uniform ground.
(a) (b)
(c) (d)
Figure 5. Electric field magnitude in the uniform ground for soil resistivity of (a) 1 Ωm, (b) 10 Ωm, (c) 100 Ωm, and (d) 1000 Ωm.
Figure 6. Vertical electric fields propagated 44 km over horizontally stratified ground. The solid lines show results calculated by the FDTD
method, and dashed lines show those by the approximated theoretical expressions.
(a) (b)
Figure 7. Time derivatives of vertical electric fields propagated 44 km over horizontally stratified ground. The solid lines show calculation result
by the FDTD method, and dashed lines show those by the approximated theoretical formula.
the FDTD method without the moving-window technique. to 500 Ωm, the electric field waveform is slightly distorted
The electric field penetrates deeper with the increase of the and the peak value is slightly decreased comparing
soil resistivity. with those in the case of a uniform ground with the soil
resistivity of 1 kΩm. However, the difference in the arrival
The ground with soil resistivity being 1 or 10 Ωm acts as
time is relatively small in terms of electric field derivative
a shield against a high frequency component such as wave waveform as shown in Figure 7 (a). When the soil resistivity
front of LEMP waveforms. The characteristic indicates that of the first layer is set to 10 Ωm, a clear difference appears
the LEMP propagation characteristics over the horizontally in comparison with the case of a uniform ground having the
stratified ground with surface resistivity of 10 Ωm or less soil resistivity of 1 kΩm. This is because the penetration
are determined by the surface. The LEMP propagation depth of the ground is about 0.5 m to 2 m at the frequency
characteristics over the stratified ground with surface having around 0.5 MHz, and the electromagnetic field does not
higher than 100 Ωm of soil resistivity are determined by the sufficiently penetrate into the second layer having a soil
parameters of the both first and second layers. resistivity of 1 kΩm.
Figure 6 and Figure 7 show examples of vertical electric Figure 8 shows the delay of LEMP arrivals in cases of
field waveforms and their time derivatives calculated by uniform and horizontally stratified ground. When the soil
the FDTD method and theoretical expression. As shown in resistivity of the first layer decreases from a relatively high
Figure 6 (a) when the soil resistivity of the first layer is set value to some tens Ωm owing to rain or some other reasons,
Figure 8. Propagation delay of lightning electromagnetic field propagating over horizontally stratified ground
observed 44 km away from the lightning strike point. The soil resistivity of the second layer is 1 kΩm.
Figure 9. Maximum value and rise time of the lightning electromagnetic field propagating 44 km over
a horizontally stratified ground.
the propagation delays are reduced from 1.0–0.9 μs to ground” (IEEE Transactions on Electromagnetic Compatibility,
around 0.4 μs. They correspond to from 300–270 m to 120 vol. 52, no. 3, pp. 657–663, 2010)
m. That means the constant propagation correction based [4] M. A. Uman, D. K. Mclain, and E. P. Krider, “The electromagnetic
radiation from a finite antenna” (American Journal of Physics,
on the soil resistivity may decrease the location accuracy of vol. 43, pp. 33–38, 1975)
the area in rainy days, which has high soil resistivity such as [5] V. Cooray, “Effects of propagation on the return stroke radiation
mountainous areas in Japan. As shown in Figure 9, though fields” (Radio Science, vol.22, no. 2, pp. 757–768, 1987)
the difference in the electric field amplitude is less than [6] V. Cooray, M. Fernando, T. Sörensen, and T. Götschl,
10%, the rise time deviates 200%. Note that the ratios are “Propagation of lightning generated transient electromagnetic
fields over finitely conducting ground” (Journal of Atmospheric
taken by setting the results derived in perfectly conducting and Solar-Terrestrial Physics, vol. 62, pp. 583–600, 2000)
ground case as the reference. [7] J. R. Wait, “The ancient and modern history of EM ground-wave
propagation” (IEEE Trans. Antennas and Propagation, vol. 40,
4. Conclusion no. 5, pp. 7–24, 1998)
[8] Y. Baba, and V. A. Rakov, “Electromagnetic computation
This paper has studied effects of ground surface conditions methods for lightning surge protection studies” (Wiley-IEEE
on propagation characteristics of lightning electromagnetic Press; 2016, 1st edition, chap. 3, pp. 43–72)
pulses, especially the horizontally stratified ground case. [9] F. Akleman, and L. Sevgi, “A novel Finite-Difference Time-
In the analysis, soil resistivity within 1 m below a ground Domain wave propagator” (IEEE Trans. Antennas and
Propagation, vol. 48, no. 3, pp. 839–841, 2000)
surface is set to a value smaller than that of the second
[10] N. Nagaoka, “Transient analysis of cable systems by means
layer, assuming the decrease of soil resistivity in an area of a frequency-transform method”(Ph.D. thesis, Department
due to precipitation. Since the ground surface with soil of Electrical Engineering, Faculty of Engineering, Doshisha
resistivity lower than some tens Ωm acts as a shielding University, 1992, chap. 3, pp. 3,19–3,24)
layer, LEMPs are not capable of penetrating into the [11] M. Saito, T. Miki, and M. Miki, “Study on development of
advanced LLS (1) –Factors of location errors of LLS–” (CRIEPI
second layer. The phenomena decrease the accuracy of the Report, Electric Power Engineering Research Laboratory, no.
propagation delay correction in rainy days at areas having H14007, 2014) (in Japanese)
high soil resistivity. The paper has shown feasibility of
propagation delay corrections depending on weather 6. Biographies
conditions for the advance of LLSs. The results obtained Akifumi Yamanaka received the B.Sc. and M.Sc.
in this paper might enable an estimation of rise-time of degrees in Electrical Engineering from Doshisha
lightning current using LLSs. University, Kyoto, Japan, in 2017 and 2019, respectively.
He is currently a Ph.D. Student at the same University.
5. Bibliographies He mainly studies lightning transients in power systems.
[1] M. Matsui, K. Michishita, S. Kurihara, and N. Honjo, Mr. Yamanaka is a student member of IEEE, the Institute
“Discussions on location accuracy improved by propagation of Electrical Engineers of Japan, and the Institute of
delay corrections for the Japanese Lightning Detection Network”
(Proc. 33rd International Conference on Lightning Protection Electrical Installation Engineers of Japan.
(ICLP) 2016, Estorill, Portugal, 2016)
Naoto Nagaoka received the B.Sc., M.Sc., and Ph.D.
[2] N. Honma, K. L. Cummins, M. J. Murphy, A. E. Pifer, and T.
Rogers, “Improved lightning locations in Tohoku region of degrees in Electrical Engineering from Doshisha
Japan using propagation and waveform onset corrections” (IEEJ University, Kyoto, Japan, in 1980, 1982, and 1993,
Transactions on Power and Energy, vol. 133, no. 2, pp. 195–202, respectively. In 1985, he joined Doshisha University,
2013)
where he has been a Professor since 1999. From April
[3] A. Shoory, A. Mimouni, F. Rachidi, V. Cooray, R. Moini, and S.
H. Sadeghi, “Validity of simplified approaches for the evaluation 2008 to March 2010, he was the Dean of the Student
of lightning electromagnetic fields above a horizontally stratified Admission Center, Doshisha University. From April
* hallas@hst.tu-darmstadt.de
KEYWORDS
High voltage testing techniques, HVDC testing, current injection, prototype installation test, prequalification test
injection is that the same current generator type can be stretched to its limits. Especially submarine cables
easily be used for the dummy loop [7]. with iron armoring will have eddy currents and hysteresis
losses in the armoring, which influences the temperature
The prototype installation test for gas insulated systems
distribution. The armoring might be removed, but the
has a different purpose and, therefore, follows a different
customer needs to accept the change in the test object.
test procedure. Gas insulated systems like switchgear or
Gas insulated systems with iron enclosures would have
lines form a more complex arrangement with several
the same limits, so that special agreements with the
contact surfaces. These are usually the most critical
customers of the HVDC equipment would have to be
components during the heating process. Therefore,
made [7].
temperature rise tests on gas insulated systems are
performed with the nominal current. The temperature Gas insulated systems shall be tested with rated current.
rise at the contact surfaces and the overall assembly is This can be achieved with rated DC current, with a
observed and has to be within the limits according to IEC representative AC current or the injection of the rated
62271-203 [3]. The same procedure is specified for gas DC current value as AC current. Testing with rated
insulated HVDC equipment [1]. Due to test equipment DC current is only possible with the use of DC current
limitations, usually AC current heating according to injection.
Figure 1 is also used for gas insulated systems. A representative AC current during the tests has to be
2.1. Technical limits of ac current testing determined by pre-tests. The customer might assume a
different temperature distribution in case of AC current
AC current heating has some technical limitations. heating instead of DC current heating, which might
Especially larger HVDC test objects of large cross distort the test results. The same or higher temperatures
sections have a higher reactive power demand due for the representative AC current compared to the
to their high inductance and high nominal current of rated DC current have to be ensured and proven to
the test object. Considering a typical current loop, the the customer. To ensure the performance for higher
overall inductance of the assembly will be in the range temperatures during testing the manufacturer of HVDC
of 1 μH/m [4]. With a testing current of 5000 A at 50 equipment has to oversize the equipment. This results in
Hz, the required reactive power would be in the range higher material investment for the design. The customer
of 7,85 kVA/m and the active power in the range of only might also assume induced currents in the overall
0,3 kW/m. Assuming a 300 m testing loop, the reactive test assembly, which might change the temperature
power would be 2.4 MVA, while the active power would distribution. For example, zero load and high load test
be in the range of only 90 kW. The need of reactive circuits in parallel will always result in induced currents
power to feed the test loop results in larger AC current in the zero load test loop [7]. To solve all these issues,
sources and power supplies. The high inductance during special agreements for testing are necessary.
AC current heating of HVDC equipment results from
the single point earthing of the outer sheath or enclosure, The injection of the rated DC current value as AC current
respectively. Single point earthing is necessary to will result in much higher temperatures at the whole test
avoid induced currents on the sheath/enclosure, which object. Therefore, the manufacturer has to oversize the
would not occur during real service operation of the total design to pass the test.
DC equipment. Such sheath/enclosure currents would
change the temperature difference across the insulation 3. DC current injection
from ΔT to ΔT’. Figure 2 shows the differences between To summarize, tests on HVDC equipment have to
two-points and single-point earthing [4]. ensure reliable operation during the total lifetime in the
HVDC application [1]. Usually power system operators
2.2. Customer arguments against ac current heating
desire a test procedure as close as possible to practice,
For special arrangements, AC current heating may also which would become possible by DC current injection
1.2
Polarity reversal test performance (polarity rev.) 3. Design and erection of the generator
1.3
DC current ripple 3.1 Overall costs of the generator
1.4
Electro-magnetic compatibility (EMC) 3.2 Use of commercial standard parts without
1.5.
Possibility of superimposed impulse voltage tests redesign (stand. parts)
(SIMP) 3.3 Weight of the overall assembly
1.6. Influence on partial discharge measurements (PD) 3.4 Upscaling / downscaling of the concept for
2. Laboratory Operation different voltage, current and power. All concepts
2.1 Handling and maintenance of the test generator of Figure 5 are compared in Table 1. Each point
2.2 Controlling of the generator is evaluated separately for each concept and is
2.3. Influences on parallel working areas marked with “+” if an advantage is estimated,
2.4. Space requirements “0” if neutral and “-” in case of a disadvantage.
1. maintenance
2. polarity rev.
2. stand. parts
2. controlling
3. Influences
4. upscaling
3. weight
5. SIMP
3. ripple
4. EMC
1. costs
4.space
6. PD
1. capacitive + 0 + - - - + + 0 + + + + +
2. hydraulic + + 0 + + + - 0 - 0 + + 0 0
3. inductive 0 0 + + 0 + + + + - - 0 + -
concept
4. mech. shaft 0 + 0 + + + - 0 - 0 0 - 0 0
5. transformer 0 0 + + + + + + + - - 0 - +
6. fuel-engine - - 0 - - 0 - 0 - + + + + 0
Table 1: Comparison of the concepts in Figure 5 (“+” = advantage, “0” = neutral, “-” = disadvantage)
during long-term DC and polarity reversal stress. Thus, the development. The feeding unit could be designed
a certain number of transformers have to be used in very simple. Only a standard regulating transformer
series, which increases costs and the weight of the overall was required. Transformers with solid insulations were
assembly. If these disadvantages can be managed, the considered and investigated during the design process.
concept will have a very stable operating behavior since Since the DC voltage has to be insulated between primary
no rotating machines are used. The concept can easily be and secondary winding, only a small insulating distance
scaled up by adding further isolating transformers. But is possible. Thus, insulation breakdowns occurred during
each transformer has to carry the rated power and thus the prototyping process. This proves that insulation faults
to be dimensioned with the required cross-section of the are more likely compared to other concepts. Figure 6 (b)
copper windings. Overall material invest and weight and shows a possible implementation of the overall assembly.
thus the costs of the design will be higher compared to It turns out that this option is far less compact than the
other concepts. capacitive solution (Figure 6 (a)).
The current injection with a fuel-power-system is a very
simple and cheap solution to fulfill the task. But the high 4. Conclusion
risk of fuel explosion in combination with electrical arcs DC current injection on high voltage potential offers
and the constant fuel feeding are crucial points for the new technical possibilities for laboratory tests. Technical
designer and the operator of such generator. The influence limits of AC current testing can be overcome. Especially
on PD measurement is uncertain [7]. the high reactive power consumption by AC heating of
3.3. Prototype experience HVDC equipment of high current ratings can be avoided.
Also all customer requirements for adequate testing
Part of the concepts described in Figure 5 have already with the real technical ratings can be fulfilled. Even
been prototyped or are actually in discussion to DC current heating during long-term tests of AC cable
overcome the limitations of AC current injection. Known systems is of interest to achieve higher current ratings
publications also show the progress in research for this and larger cross sections. Several technical solutions
new generator type. Figure 6 shows CAD prototypes of can be developed to inject DC current on high voltage
different generator concepts. potential. Each technical solution has its advantages
A capacitive current injection generator is being built up and disadvantages. Running prototyping projects for the
in the high voltage laboratories at Technische Universitaet generator concepts are shown. Especially the capacitive
Darmstadt. First experiences with smaller prototypes current injection seems very promising and is currently
were positive and showed good performance. The followed.
construction of a larger generator for 5000 A and 660 kV
voltage is currently in progress. It consists of a feeding 5. Acknowledgment
unit on ground potential, two capacitor columns similar to The authors gratefully acknowledge support of this
voltage dividers and the rectifier on high voltage potential. work by the IWB-EFRE-Program of the State of Hessen
Due to the high frequency of approximately 50 kHz the (Funding Code 20002558).
assemblies on high voltage potential can be designed very
compact.
The current injection with isolating transformers had
been investigated before, but was discarded during
* Konstantinos.velitsikakis@dnvgl.com
** siukilau@clp.com.hk
KEYWORDS
Line arresters, EGLA, Lightning performance, Shielding failure, Backflashover
Hong Kong is located in the sub-tropical zone having density (GFD, number of flashes/km2/year) is estimated
fairly high lightning activity. With the increasing use for various locations in Hong Kong, as presented in
of electronic controls for motor drives in industrial and Figure 2a. In the study, the GFD was considered equal
commercial sectors, power quality, in particular voltage to 5.2 flashes/km2/year, corresponding to the average
dips, is becoming a prominent issue. CLP is committed GFD, as recorded in 2016.
to continuously improve the services to customers of A typical configuration of a 400 kV suspension tower
Hong Kong and, therefore, has initiated the installation in CLP overhead transmission lines is shown in Figure
of line arresters on the 400 kV overhead transmission 3. In addition, for those phases of the lines protected by
lines since early 2000s. Following the increased line means of EGLAs, there is a fixed external gap between
arrester coverage on the 400 kV overhead transmission the arresters and the phase conductor (Figure 4).
lines, the lighting performance of the latter has been
significantly improved. The local terrain features for each tower structure were
deemed essential for studying the lightning performance
In 2017, CLP conducted a detailed study to evaluate the of an overhead transmission line against shielding failure
effectiveness of installing additional line arresters on the flashovers. In the study, the terrain features referred to
400 kV double-circuit overhead transmission lines. The the ground elevation and the ground inclination angle,
purpose of the study was a) to evaluate the effectiveness φ°, at the location of each tower. An example is given in
of increasing the total number of installed arresters on Figure 5 that presents the terrain features along a 400 kV
the lightning performance of the lines and b) to define line in Hong Kong consisting of 24 towers (tower No.1
cost-effective application strategies. refers to the substation gantry). The analysis of the
terrain features for all lines have shown that in average
2. Overview of the input data 54% of the tower structures are erected on a flat terrain,
CLP owns and makes use of a Lightning Location System i.e. φ<5°, 40% are erected on a moderately hilly terrain,
(LLS) to record lightning activity in Hong Kong. Based i.e. 5° ≤φ<15°, and 6% are erected on an extremely hilly
on these recordings, the yearly average ground flash terrain, i.e. φ≥15°.
Figure 2a. GFD map in Hong Kong Figure 2b. Average recorded GFD in Hong Kong
Figure 5. Terrain features of the tower structures of a 400 kV double-circuit line in Hong Kong
Figure 6. EGLA schemes (filled square in red indicates the presence of an arrester at that phase)
The analysis showed that the backflashover performance the lightning current results in a voltage built-up on the
is to a large extent influenced by the tower footing metallic structure as well as on the shield wire. Each
resistance. This is because of its direct impact on the phase conductor is electrically coupled to the shield wire
minimum critical lightning current that leads to a and the coupling factor is distance dependent. Therefore,
flashover of at least one phase. More specifically, for the top phase conductor, which is located to a closer
tower structures in areas where low values of the tower distance to the shield wire, will experience a higher
footing resistance can be achieved (e.g. Rtower≤10 ohms), coupled voltage compared to the middle and bottom phase
the minimum critical currents have extremely high respectively. Consequently, the voltage stress across the
values, resulting in very low backflashover failure rates. line insulation is lower for the top phase and higher for
The latter is illustrated in Figure 7a, where the critical the bottom phase. Thus, by protecting the bottom phase
lightning current is greater than 230 kA for each phase by means of a line arrester, it was calculated that the
angle of the power frequency voltage. On the other hand, backflashover rate would be significantly reduced.
for tower structures in areas where higher tower footing
resistances occur (e.g. Rtower >10 ohms), lower critical
4. Shielding failure flashover
currents occur, resulting in higher backflashover rates. analysis
For these rogue structures, the study results revealed According to reports published by Utilities in China
that the bottom phase of a circuit is more prone to [6,7], it has been observed that the shielding failure
backflashovers, followed by the middle and the top rates in their transmission lines are higher compared to
phases respectively. This can be explained as follows: the estimated rates, as calculated by the conventional
Line 1 50 50 0
Line 2 25 75 0
Line 3 50 35 15