Research Proposal

JOMO KENYATTA UNIVERSITY OF AGRICULTURE AND TECHNOLOGY
PhD Research Proposal

An Investigation into the Applicability of Harmony Search Algorithm in Optimal Load Shedding
MWANIKI CHARLES
5/1/2013
TABLE OF CONTENTS
1 INTRODUCTION .............................................................................................................. 3 1.1 1.2 1.3 2 Relevance of Study/Justification ................................................................................. 3 Problem Statement ....................................................................................................... 4 Objectives .................................................................................................................... 5
LITERATURE REVIEW ................................................................................................... 6 2.1 2.2 Under voltage Load Shedding ..................................................................................... 6 Under frequency Load Shedding ................................................................................. 8
2.3 Under voltage under frequency Load Shedding .......................................................... 15 3 4 5 6 METHODOLOGY ........................................................................................................... 17 TIME FRAME .................................................................................................................. 18 BUDGET .......................................................................................................................... 19 REFERENCES ................................................................................................................. 20
1 1.1
INTRODUCTION Relevance of Study/Justification
The increase in electrical power consumption is directly proportional to the increase in people population. As of 10th July, 2009, the world population was estimated by the United States Census Bureau at 6.77 billion. This figure is expected to reach about 9 billion by the year 2040 [2]. Many environmental and economic constraints prevent the construction of new or upgrading of the existing generation and transmission capacities. Additionally, generation reserves are minimal and often the reactive power is insufficient to satisfy the load demands.Given this trend, power systems are expected to be more heavily loaded and moving closer and closer to stability limit and more susceptible to disturbances and outages.
Blackouts of power systems always have been a historical problem in interconnected power systems. However in recent years by improving monitoring and protection techniques, it is not possible to completely prevent of blackouts [1-3]. Sudden and large changes in generation capacity such as the outage of a generator can produce a sever imbalance between generation and load demand. This may lead to a rapid decline in frequency, because the system may not respond fast enough. If voltage and frequency get out from permissible range the system is in unstable condition. In this condition the system controller's operate and attempt to restore the voltage and frequency in the permissible range. If the disturbance is so large the controller's cant restore the voltage and frequency in the permissible range. In this condition the last solution to avoid the power system breakdown has been load shedding strategy.
Recent blackouts have brought our attention to the issues of voltage stability in the system. Voltage decline can be a result of a disturbance. Its main cause, however, is insufficient supply of reactive power. This has led researchers to focus on techniques to maintain voltage stability. The loss of a generator causes an unbalance between the generated power and the load demand. This affects the frequency and voltage. Load shedding schemes must consider both these parameters while shedding load. By shedding the correct amount of load from the appropriate buses, the voltage profile at certain buses can be improved [4].
While considering the amount of load to be shed and the step size, it is also important to take into account the reactive power requirements of each load. Quite often, disturbances such as a generator loss cause the voltage to decline. An effective way to restore voltage is to reduce the
reactive power demand. Thus when loads absorbing a high amount of reactive power are first shed; the voltage profile can be improved.
1.2
Problem Statement
Load shedding is an emergency control action to ensure system stability, by curtailing system load. The emergency load shedding would only be used if the frequency or voltage falls below a specified frequency/voltage threshold. Typically, the load shedding protects against excessive frequency or voltage decline by attempting to balance real and reactive power supply and demand in the system. The number of load shedding steps, amount of load that should be shed in each step, the delay between the stages, and the location of shed load are the important objects that should be determined in a load shedding algorithm. Despite being successful to a great extent, the conventional load shedding schemes have certain disadvantages. The amount of a load step is, at times, large which causes excessive load to be shed. Most schemes do not have the flexibility to increase the number of load shedding steps, thereby introducing transients in the system. The most LS schemes proposed so far used voltage and frequency parameters, separately and also, the under-frequency and under-voltage relays are working in the power system without any coordination. The individual use of these indices may be also not reliable/effective, and may even lead to the over load shedding problems. Studying on the under-frequency load shedding is often done using the system frequency response models. The impact of voltage variation on the frequency deviation is not considered in these models. Furthermore, the UVLS methods that are proposed so far for adjusting the under-voltage relays, does not consider the frequency behavior. These two parameters (voltage and frequency) are not independent and the coordination between UFLS and UVLS schemes is therefore crucial. The dependency between voltage and frequency will affect LS performance. Economical considerations need to be considered before shedding the load since certain loads cannot be kept offline. Further, Load shedding is an emergency control operation and should be on a priority basis, which means shedding less important loads, while expensive industrial loads are still in service. Therefore, an algorithm that uses voltage and frequency parameters simultaneously for making load shedding decisions on a priority basis, will be more reliable and effective than the conventional schemes.
This study focuses on establishing the applicability of Harmony Search algorithm in optimal load shedding with a view to overcome these issues. This will help in power system planning, operation and control.
1.3
Objectives
The problem, to be treated in this study, will have three objectives: a) Coding of harmony search algorithm subroutine for use in PSS/E software b) Determine the applicability of harmony search algorithm in under frequency load shedding scheme c) Determine the applicability of harmony search algorithm in under voltage load shedding scheme
LITERATURE REVIEW
Different methods for load shedding and restoration have been developed by many researchers. Currently there are various under frequency and under voltage load shedding techniques used in the power industry worldwide.
2.1 Under voltage Load Shedding Lopes et all [5] suggests a method which carries out load shedding in case of two conditions. One, where the load shedding occurs due to a post disturbance low voltage condition and secondly, where the load shedding results due to the inability of the system to achieve a stable operating condition during post disturbance. This method uses the load flow in order to decide the buses from which to shed load. The initial set of control actions are first carried out. These actions are capacitor switching, tap changing transformer and secondary voltage control.
Jianfeng et al [6] have developed a method with risk indices in order to decide which buses should be targeted for load shedding to maintain voltage stability. The buses with a high risk of voltage instability are considered first. This is estimated from the probability of a voltage collapse occurrence. The risk indices are the products of these probabilities and impact of voltage collapse.
Another method [7][31] dealing with the particle swarm approach for under voltage load shedding has been researched. The particle swarm Optimization concept is a group or cluster of particles in which each particle is known to have individual memory like an animal in its herd or flock. The flock is initiated with some initial velocity and the particles move in different directions to come up with the best solution. The best solution is shared with every particle of the group so that they can move from there on based on this new acquired knowledge. This same idea is used for under voltage load shedding to recognize the best possible load shedding scheme considering the system conditions and disturbance particular to that situation.
Ladhani and Rosehart [8] propose load modeling for an under voltage load shedding scheme. They also suggest offering economic incentives to customers for discontinuing the use of power during load control periods. This way the brunt of a sudden load shed is not borne by the customer alone. Also, systematic load control will lead to the stability of the system even when it is not faced with a disturbance. 6
Yorino et al [9] suggests a new planning method for planning the VAR allocation using the FACTS devices. Here, the total economic cost for a voltage collapse along with its corrective control and load shedding are taken into account to come up with the optimum VAR planning scheme. Thus, the objective function is to minimize the cost while keeping in mind the voltage stability of the system.
Mozino [10] discusses the currently existing under voltage load shedding schemes. They are divided into two categories; decentralized and centralized. The decentralized load shedding involves setting relays at buses with loads to be shed and tripping the respective relays. The centralized scheme is more advanced. The relays are installed at the key bus locations and the information regarding which relays are to be tripped is sent to these relays from a main control centre. Thus the required load is shed from appropriate buses. Many of these schemes are referred to as special protection or wide area schemes. The two categories mentioned above are widely used as under voltage load shedding relays. These relays require logic and have to perform efficiently and accurately. Also, these relays must avoid false operation. Thus to satisfy the above requirements digital relays are being used for under voltage load shedding.
Single Phase UVLS Logic measures voltages on every phase. This scheme distinguishes between voltage collapse and fault induced low voltages. The voltage collapse is a balanced phenomenon, hence results in a reduction of voltage on all the three phases. Except for a three phase fault all the other faults are unbalanced. The relays trips when it identifies a voltage collapse and blocks the relay for a fault induced low voltage. Unbalanced faults usually induce negative sequence voltages which are detected and used for blocking the relay. Positive sequence UVLS logic checks the positive sequence voltage with the set point value. Since the voltage collapse is balanced for all the three phases, the positive sequence voltage is equal to the three phase voltages. In case of a fault condition, the negative sequence voltage is utilized to block the relay.
A load shedding scheme against long term voltage instability is proposed by Van Cutsem et al [11]. It uses distributed controllers which are delegated a transmission voltage and a group of loads to be controlled. Each controller acts in a closed loop, shedding loads that vary in magnitude based on the evolution of its monitored voltage. Each controller acts on a set of 7
electrically close loads and monitors the voltage V of the closest transmission bus in that area. The controller is rule based where the rules are simple if-then statements.
2.2
Under frequency Load Shedding
Under frequency load shedding mainly sets up relays to detect frequency changes in the system. As soon as the frequency drops below a certain value a certain amount of load drops, if the frequency drops further, again a certain amount of load is dropped. This goes on for a couple of steps. The amount of load to be shed and the location of the load to be shed is predetermined. Terzia [12] talks about under frequency load shedding in two stages. During the first stage the frequency and rate of frequency changes of the system are estimated by non-recursive Newton-type algorithm. In the second algorithm, the magnitude of the disturbance is estimated using the simple generator swing equation.
In another approach Thalassinakis et al [13] obtained results from an autonomous power system on the Greek Islands of Crete. The method uses the Monte Carlo simulation approach for the settings of load shedding under frequency relays and selection of appropriate spinning reserve for an autonomous power system. The settings of the under frequency relays are based on the four parameters; the under frequency level, rate of change of frequency, the time delay and the amount of load to be shed. Three sets of system indices are defined. These sets are for the purpose of comparisons between load shedding strategies. A method was developed which simulated the behavior of a power system. The three aspects of the power systems that were developed in the simulation were Operation of the power system as performed by the control centre. Primary regulation of the generating units after the failure of a generating unit. Secondary regulation and utilization of the spinning reserves.
Three different cases of comparing the spinning reserves with the load mismatch are considered. One, when the spinning reserve is sufficient or greater. Thus the load can be restored immediately. Second, when the spinning reserve is slightly insufficient and the rapid generating units will require a certain amount of time to be started. Thus it will be 10-20 minutes before the load can be completely restored. Third, the spinning reserves are insufficient and there are not enough rapid generating units thus implying that the load will not be restored for a considerably long period of time.
Another method [14] triggers the under frequency relays based on a dynamically changing intelligent load shedding scheme. The main components of this scheme are the knowledge base, disturbance list and the ILS computation engine.
Fig. 1 Block Diagram of the ILS scheme
The generalized structure of the ILS scheme is shown in figure 1. The knowledge base is the most important block. It is connected to the computation engine which sends trip signals to relays. The network models can be accessed by the knowledge base while monitoring the system. The knowledge base is trained and its output consists of system dynamic scenarios and frequency responses during disturbances. This trained knowledge base also monitors the system continuously for all operating conditions. The disturbance list consists of pre-specified system disturbances. Based on the inputs for the system and the continuous system updates, the knowledge base notifies the ILS engine to update its load shedding list. Thus it ensures that the load shed is always minimum and optimum.
Wee-Jen Lee [15] discuss about another intelligent load shedding based on microcomputers. The unique feature about this scheme is the built in frequency setting and the time delay setting. The frequency setting in the relay counters system re collapse situation. (Consider a generator loss which triggers a load shedding step. This causes the frequency of the system to recover. During this recovery period if another generator trips it results in a system re collapse). Typical frequency relays will not trip until the second generator loss causes sufficient frequency decay. The ILS system automatically adjusts the frequency settings such that load is shed immediately without delay. The time delay settings cause the load scheme to initiate during situations when a disturbance causes the frequency to drop and hold at a value less than the rated. The number of load shedding steps can be increased without a limit. The advantage of having large number of 9
load shedding steps is that it prevents large amount of transients. It also prevents over shedding. Denis Lee Hau Aik, [16] suggests a method using the System Frequency Response SFR and the Under Frequency Load Shedding UFLS together to get a closed form expression of the system frequency such that the UFLS effect can be included in it. On doing this, the system and UFLS performance indicators can be calculated. Thus these indicators can be used efficiently in any further optimization techniques of SFR UFLS model. One such method has been discussed using the regression tree by Chang et al [17]. The regression tree is utilized to interpolate between recorded data to give an estimate of the frequency decline after a generator outage. It is a non parametric method which can select the system parameters and their relations which are most relevant to the load imbalance (due to generator outage) and the frequency decline. The case considered here is only a generator outage but this method can be applied to other forms of disturbances as well.
A Kalman filtering-based technique by A.A. Girgis et al [18] estimates frequency and its rate of change which is beneficial for load shedding. The noisy voltage measurements are used to estimate the frequency and its rate of change. A three-state extended Kalman filter in series with a linear Kalman filter is used in a two stage load shedding algorithm. The output of the three stage Kalman filter acts as the input to the linear Kalman filter. It is the second filter which identifies linear components of the frequency and its rate of change. The amount of load to be shed is calculated using the linear component of the estimated frequency deviation.
Another method uses Kalman filtering [19] to estimate the frequency and its rate of change from voltage waveforms. The buses are ranked based on their rate of change of voltage (dV/dt) values. The disturbance magnitude is calculated from the swing equation. The rate of change of frequency required for this equation is calculated using the Kalman filter. Once the total amount of load to be shed is estimated then the load to be shed from each bus is determined based on the PV analyses.
An optimization technique for load shedding [20] with distributed generation was developed. This technique converts differential equation into algebraic ones using the discretization method. Two cases are considered here; one with the distributed generation switched on to the system as a static model and the other case without the distributed generation on the grid. Both cases resulted in successful shedding of appropriate quantity of load. 10
Li Zhang suggests a method [21] which designs under frequency relays using both the frequency and the rate of change of frequency (df/dt). The scheme has been designed for a 50 Hz Northeast China power system. Traditional schemes required only the frequency decay information. Here the rate of change of frequency is used as auxiliary information. The plots for the rate of change of frequency are oscillatory in nature. Hence a new scheme is devised in this paper which considers the integration of the rate of change of frequency (df/dt) to indicate the frequency drop. By integrating one is effectively measuring the area between two frequencies, fi-1 and fi. The schemes is made up of five load shedding steps for a 50 Hz system. These steps are from 50 to 49.2 Hz, 49.2 to 49 Hz, 49 to 48.8Hz, 48.8 to 48.6 Hz, 48.6 to 48.4 Hz. The amount of load to be shed in each step is decided by integrating the df/dt value in each step. The simulation results when compared with the old scheme with just the frequency decay show a definite improvement in system frequency due to the inclusion of rate of change of frequency (df/dt) in the new scheme.
The main idea in the paper proposed by Xiong et al [22] is the inclusion of on line load frequency regulation factors. Loads with smaller frequency regulation factors are shed first, followed by the ones with larger frequency regulation factors. The active power and load frequency relation is established in the form of the following equation.
Where, fN is the nominal frequency. PLN is the rated active power and ai (i=1,2n) is the percentage of the total load associated with the i-th term of the frequency. The per unit form of the above equation is differentiated to get the change in load power as frequency changes (dPL/df) which is the KL factor or regulation factor. The higher order terms are neglected.
Thus it is preferable to shed load for smaller regulation factors. Hence the loads are distinguished based on their individual regulation factors and accordingly load shedding schedules are planned based on their respective K factors.
11
Another scheme considering the rate of change of frequency is the adaptive load shedding algorithm in the paper by Seyedi et al [23]. Here the shedding is adapted as per the intensity of the disturbance. This intensity is determined based on the rate of change of frequency. Thus the main points observed while designing the scheme is that the speed of load shedding is increased if the rate of change of frequency (df/dt) values are high. Also, the number of load shedding steps and the amount of load to be shed in each step is increased if there is an increase in the rate of change of frequency (df/dt) values. The new method was tested on the HV network of the Khorasan province in Iran. The proposed method definitely showed improvements as compared to the conventional scheme.
Neural networks are proposed [24] to be used for an under frequency load shedding scheme. This intends to replace the conventional slow acting dynamic simulators by quick and efficient neural network engines. The general procedure is to identify the inputs for the neural networks, generations of data sets, designing NN and the evaluating the performance of neural nets. The variables used as inputs are the actual real power generation, available real power, actual load generation level prior to a disturbance, amount of the actual load being shed and the percentage of the exponential load to be shed.
A SCADA based scheme has been proposed by Parniani et al [25]. The rate of change of frequency is useful in identifying the overload when a disturbance occurs and hence is helpful to estimate the amount of load to be shed. The SCADA based scheme overcomes the shortcomings of the previous adaptive UFLS scheme. The mean system frequency is defined as follows,
Where fi is the frequency of the generators from 1 to n and H is their respective system inertia. Adding the df/dt equation every generator the post disturbance equation obtained for SCADA is
12
where
PL is the disturbance magnitude in per unit. Now another variable
Pthr is defined. If
a disturbance occurring at the weakest generator is less than this value then absolute frequency of that generator is within the permitted limits. For a situation where the isturbance magnitude, PL is less than Pthr no load shedding is required. The maximum load shedding Pthr ( PL -
magnitude is equal to the difference between the disturbance magnitude and
Pthr ). The load to be shed is distributed inversely proportional to the generator inertia to make the load shedding most effective. The equation (4) represents this distribution.
Based on this equation the layers of the load shedding scheme are designed. Both the steps shed one third of the remaining load. These are in steps. They are presented in a table 1 with the first step being at 59.3 Hz.
TABLE 1: SCADA Based Load Shedding Formula
13
An adaptive load shedding scheme which includes a self healing strategy is presented by Vittal et al [26]. The proposed scheme is tested on a 179 bus 20 generator test system. This self healing strategy comes into play when the system vulnerability is detected. The system then divides into self sustaining islands. After this islanding, load shedding based on the rate of change of frequency is applied to the system. Due to this division, it becomes easier to restore load. A Reinforcement Learning scheme is discussed in the paper. The first is the controlled islanding which is done using the two-time scale method. It deals with the structural characteristics of the power systems and determines the interactions of the generators and their strong or weak coupling. The Dynamic Reduction Program 5.0 (DYNRED) is the software in which simulations are run to implement this technique. Through this software coherent group of generators can be obtained on the power system. Islanding causes two types of islands to be formed, the generation rich islands and the load rich islands. The load rich islands may have a further decline of frequency. This may result in the generator protection to trip the generators thus further declining the islands frequency. Thus a two layer load shedding strategy is employed for the load rich island. The first layer is based on the frequency decline approach. The second layer considers the rate of change of frequency. Due to the longer time delays and lower frequency thresholds for a frequency based scheme inadvertent load shedding is avoided. When the system disturbance is large and exceeds the signal threshold, the second layer comes into play. It sends a signal to discontinue the first layer of operation and continues with the load shedding based on rate of change of frequency. This layer will shed more load at the initial steps to prevent cascading effects. The magnitude of the disturbance is found based on the formula
If we sum up all the equations for i=1 to n then the final equation obtained is
Where, m0 is defined as df/dt which is the average rate of frequency decline. Rearranging the above equation we get a new equation which relates PL to m0 . 14
Since Hi is constant, the magnitude of m0 can be directly proportional to the rate of frequency decline. Hence the rate of change of frequency df/dt ) can be a measure of the disturbance. Once the disturbance threshold value, PL , for the second layer of load shedding is decided, the m0 value is calculated. The mi at each bus is calculated and compared with m0 . If mi m0 then the second layer is activated, otherwise the conventional load shedding scheme is used. This new shedding scheme increases the stability of the system by shedding fewer loads as compared to the conventional scheme.
Application of Neural Network in load shedding and some Predictable functioning of load shedding methods has been proposed [28 -30][32]. In this method the identification of the variables like inputs and outputs is an important step for a successful application of this technique. Sometimes a pre-processing stage is needed to choose the most significant variables to be used as inputs of a NN. Some of the meaningful variables that have been used as inputs of the NN Active real power generation Active load generation Amount of active load being shed Percentage of exponential type loads being shed Damping factor Power factor
These variables provide the NN with valuable information, such that it can make the required assessment with respect to how much the generation load disproportion has been corrected and the influence each load type has on the resulting frequency response. 2.3 Under voltage under frequency Load Shedding
A load shedding scheme that incorporates, the frequency and the bus voltages, for deciding the instant, the amount and the location of the load to be shed is proposed [27]. The scheme developed consists of a stepwise approach. This has been represented in the form of an algorithm.
15
The first step of the load shedding procedure is the measurement and calculation of the rate of change of frequency. Depending on the relay, the frequency measurements or the rate of change of frequency are recorded in the system. The total load mismatch between the generated power and load power is determined. For a single machine, the swing equation [1] is given by
where, f0 is the nominal frequency of the system and Pdiff is the difference in the generated power and the load power. In the above equation is replaced by f since = 2f . Thus a relation between the frequency and the power mismatch is obtained. This relation establishes the estimated magnitude of the disturbance. The inertia constant in the above equation is the kinetic energy Wk over the system base MVA. The inertia constants of all the machines in the system are on the base MVA. In a large power system where there are many generators which maybe geographically far away from each other, Heq = individual inertias of each machine for all the machines in the system Also, the equivalent mechanical and electrical powers are given as;
Pm = individual mechanical shaft power of each machine for all the machines in the system Pe = individual electrical power of each machine for all the machines in the system
Once the magnitude of the disturbance is determined using the above equivalent swing equation, the location and the amount of load to be shed from each bus has to decided. In order to do this, the buses are ranked according to the dV/dt values at the point of detection of frequency decline. The bus with the largest dV/dt is listed at the top of the list and then so on in the decreasing order. Once the order is decided, the next step is to decide the amount of load to be shed at each bus. This is decided based on the voltage sensitivity at each bus. Thus the bus with voltage sensitivity very close to the instability limit will have a maximum load shed based on the reciprocal of its sensitivity as a fraction of the sum of the reciprocals of all the load bus sensitivities.
16
The load shedding scheme has been tested on the IEEE 39 bus and IEEE 145 bus test systems. Three studies based on contingency were carried out with each contingency being considered at a time: Case study 1: Loss of a generator for the IEEE 39 bus system. Case study 2: Loss of a generator for the IEEE 145 bus system. Case study 3: Loss of a transmission line IEEE 39 bus system
The scheme is simple and does not involve complicated calculations. It proved to be successful in restoring the frequency within its pre-defined limits. It has also improved the voltage profile at certain buses which had critically low voltage before load shedding was applied. However the algorithm did not address some aspects and therefore requires fine tuning. Economical considerations need to be considered before shedding the load since certain loads cannot be kept offline. The study and testing of the scheme in a multiple contingency scenario like loss of a generator along with a loss of transmission line need to be considered as this would create a critical situation.
Thus the various conventional schemes, under frequency schemes and under voltage load shedding schemes have been discussed above. These give an insight about the technological advancement achieved in this area. The proposed study intends to investigate the applicability of harmony search algorithm as an optimization tool and how it can be applied to address the disadvantages faced by the conventional schemes present in the industry.
METHODOLOGY
17
TIME FRAME
18
BUDGET
19
6 [1]
REFERENCES D. Craciun, S. Ichim, and Y. Besanger, A new soft load shedding: Power system stability with contribution from consumers, IEEE/PTC, pp. 1-6, June/July 2009. D.Andersson, P.Elmersson, A.Junti, Z.Gajie, D.Karlsson, and L.Fabiano, Intelligent load shedding to counteract power system instability, IEEE/PES, pp. 570-574, 2004. G. Shahgholian, M. E. Salary Effect of Load Shedding Strategy on Interconnected Power Systems Stability When a Blackout Occurs, International Journal of Computer and Electrical Engineering, Vol. 4, No. 2, pp 212-217, April 2012 B. Suneetha, S. Hemachandra, B. Gowri Prasad and A. Jayanth, Application of Neural Network in Load Shedding and Some Predictable Functioning of Load Shedding Methods, International Journal of Electronics, Communication & Instrumentation ngineering Research and Development (IJECIERD), Vol. 3, Issue 2, pp 45-50, Jun 2013
[2]
[3]
[4]
[5]
B. Isaias Lima Lopes, A.C. Zambroni de Sow, (2003) An Approach for UnderVoltage Load Shedding, IEEE Bologna PowerTech Conference.
[6]
Dai Jianfeng, Zhou Shuangxi, Lu Zongxiang, A New Risk Indices Based Under Voltage Load Shedding Scheme.
[7]
T. Amraee, B. Mozafari, A.M. Ranjbar, An Improved Model for Optimal Under Voltage Load Shedding: Particle Swarm Approach.
[8]
Shamir S. Ladhani, William Rosehart, Criteria for Load Control when Considering Static Stability Limits.
[9]
Naoto Yorino, E. E. El-Araby, Hiroshi Sasaki, Shigemi Harada, A New Formulation for FACTS Allocation for Security Enhancement Against Voltage Collapse.
[10] [11]
Charles J. Mozina (Beckwith Electric Co., Inc.), Undervoltage Load Shedding. Mark Adamiak, William Premerlani, Dr. Bogdan Kasztenny, Synchrophasors: Definition, Measurement, and Application.
[12]
[21] Vladimir V. Terzija, (2006) Adaptive Underfrequency Load Shedding Based on the Magnitude of the Disturbance Estimation, IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3.
[13]
Emmanuel J. Thalassinakis, Member, IEEE, and Evangelos N. Dialynas,(2004)
20
A Monte-Carlo Simulation Method for Setting the Underfrequency Load Shedding Relays and Selecting the Spinning Reserve Policy in Autonomous Power Systems, IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 19,NO. 4. [14] Farrokh Shokooh, Shervin Shokooh, Jacques Tastet, Hugo Castro, Tanuj Khandelwal, Gary Donner, An Intelligent Load Shedding System Application in a Large Industrial Facility. [15] [16] Wen-Jen Lee, Microcomputers based intelligent load shedding. Denis Lee Hau Aik, Member, IEEE, (2006) A General-Order System Frequency Response Model Incorporating Load Shedding: Analytic Modeling and Applications, IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21,NO.2. [17] Rung-Fang Chang, Chan-Nan Lu, Tsun-Yu Hsiao, (2005) Prediction of Frequency Response After Generator Outage Using Regression Tree, IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 20, NO. 4. [18] Adly. A. Girgis, William Peterson, Optimal estimation of frequency deviation its rate of change for load shedding. [19] Adly.A.Girgis, Shruti Mathure, Dissertation on Load shedding using voltage, frequency and their rate of change. [20] Adly.A.Girgis, Ding Xu, Optimal Load shedding strategies in power systems with distributed generation. [21] [22] Li Zhang, Jin Zhong, UFLS Design by using f and integrating df/dt. Xiaofu Xiong, Wenyuan Li, A New under frequency load shedding scheme considering load frequency characteristics. [23] H. Seyedi, M. Sanaye-Pasand, M.R. Dadashzadeh, Design and Simulation of an adaptive Load Shedding Algorithm using a real network. [24] Matthew A. Mitchell, J.A. PeGas Lopes, J.N. Fidalgo James D. McCalley, Using a Neural Network to Predict the Dynamic Frequency Response of a Power System to an Under-Frequency Load Shedding Scenario. [25] M. Parniani, A. Nasri, SCADA based under frequency load shedding integrated with rate of frequency decline. [26] Juhwan Jung, Chen-Ching Liu, Steven Tanimoto, Vijay Vittal, Adaptation in Load Shedding under vulnerable operating conditions. [27] Poonam M. Joshi, Load Shedding Algorithm using voltage and frequency data, 2007 and
21
[28]
B. Suneetha, S. Hemachandra, B. Gowri Prasad & A. Jayanth, Application Of Neural Network in Load Shedding and some Predictable Functioning of Load Shedding Methods, International Journal of Electronics, Communication & Instrumentation Engineering Research and Development (IJECIERD) Vol. 3, Issue 2, Jun 2013, 45-50 E. J. Thalassinakis, E. N. Dialynas, D. Agoris, Method Combining Ann's and Monte Carlo Simulation for the Selection of the Load Shedding Protection Strategies in Autonomous Power System, IEEE Trans. on Power System, Vol. 21, No.4, pp. 157482, 2006. C. T. Hsu, M. S. Kang and C. S. Chen, Design of Adaptive Load Shedding by Artificial Neural Networks, IEE Generation, Transmission, Distribution, Vol. 152, No. 3, pp. 415-421, 2005. T. Amraee, B. Mozafari, A. M. Ranjbar, An Improved Model for Optimal Under Voltage Load Shedding, Particle Swarm Approach, IEEE Power Conf. India, April 2006. Gh. Isazadeh, R. Hooshmand and A. Khodabakhshian, Design Of An Adaptive Dynamic Load Shedding Algorithm Using Neural Network in the Steelmaking Cogeneration Facility, IJST, Transactions of Electrical Engineering, Vol. 36, No. E1, Pp 67-82,2012
[29]
[30]
[31]
[32]
A METAHEURISTIC LOAD SHEDDING ALGORITHM USING VOLTAGE AND FREQUENCY PARAMETERS
CHARLES MWANIKI
22
A research proposal submitted to the Faculty of Engineering in Partial fulfillment of the requirements for the Doctor of Philosophy in Electrical Engineering Jomo Kenyatta University of Agriculture and Technology.
J.K.U.A.T July, 2013
23
DECLARATION
I declare that this research proposal is my original work and it has not been presented for an award of a degree, diploma or certificate in this or any other university.
CHARLES MWANIKI
SIGN:_________________DATE: _____________
RECOMMENDATION/APPROVAL
This research proposal has been submitted with my approval as the university supervisor.
DR CHRISTPHER MAINA MURIITHI SIGN_______________DATE: ________
This research proposal has been submitted with my approval as the university co-supervisor.
DR NICODEMUS ABUNGU
SIGN___________________DATE: ______________
24
TABLE OF CONTENTS
DECLARATION ..................................................................................................................... 24 RECOMMENDATION/APPROVAL ..................................................................................... 24 TABLE OF CONTENTS ......................................................................................................... 25 1 INTRODUCTION ............................................................................................................ 26 1.1 1.2 1.3 1.4 Background ................................................................................................................ 26 Problem Statement ..................................................................................................... 26 Relevance of Study/Justification ............................................................................... 27 Objectives .................................................................................................................. 28 Main objective .................................................................................................... 28 Specific objectives .............................................................................................. 29
1.4.1 1.4.2 1.4 1.5 2
Scope of Study ........................................................................................................... 29 Importance of the study ............................................................................................. 29
LITERATURE REVIEW ................................................................................................. 30 2.1 2.2 Under voltage Load Shedding ................................................................................... 30 Under frequency Load Shedding ............................................................................... 32
2.3 Under voltage under frequency Load Shedding .......................................................... 39 3 METHODOLOGY ........................................................................................................... 42 3.1 3.2 3.3 4 5 6 Metaheuristics algorithm ........................................................................................... 42 Harmony Search Algorithm....................................................................................... 42 Cuckoo Search Optimization ..................................................................................... 43
THESIS DEVELOPMENT SCHEDULE......................................................................... 45 BUDGET .......................................................................................................................... 46 REFERENCES ................................................................................................................. 47
25
7 7.1
INTRODUCTION Background
The developing industries and their growing infrastructure have stressed the power industry to supply sufficient power. The generation capacity should increase in proportion to the increase in the number of loads. Large power transfers across the grid lead to the operation of the transmission lines close to their limits. Additionally, generation reserves are minimal and often the reactive power is insufficient to satisfy the load demands. Due to these reasons power systems become more susceptible to disturbances and outages. Some of the disturbances experienced by the power system are faults, loss of a generator and or transmission line and sudden switching of loads [1-3]. These disturbances vary in their intensity. As a result it is necessary to study the system and monitor it in order to prevent it from becoming unstable. The two most important parameters to monitor are the system voltage and frequency, both of which must be maintained within prescribed limits standards to ensure that the system remains stable. The frequency is mainly affected by the active power, while the voltage is mainly affected by the reactive power. Specifically, the frequency is affected by the difference between the generated power and the load demand. This difference is caused due to disturbances which reduce the generation capacity of the system. For example, due to the loss of a generator, the generation capacity decreases while the load demand remains constant. If the other generators in the system are unable to supply the power needed, then the system frequency begins to decline. To restore the frequency within the prescribed limits a load shedding scheme is applied to the system. In addition, the reactive power demand of the load affects the voltage magnitude at that particular bus. When the power system is unable to meet the reactive power demands of the loads, the voltages become unstable. In such situations, capacitor banks are switched on to supply the reactive power to the loads. However, when these capacitor banks are unable to restore the voltage levels within their upper and lower limits, the system resorts to load shedding.
7.2
Problem Statement
Load shedding is an emergency control action to ensure system stability, by curtailing system load. The emergency load shedding would only be used if the frequency or voltage falls below a specified frequency/voltage threshold. Typically, the load shedding protects against excessive frequency or voltage decline by attempting to balance real and reactive power
26
supply and demand in the system. The number of load shedding steps, amount of load that should be shed in each step, the delay between the stages, and the location of shed load are the important objects that should be determined in a load shedding algorithm. Despite being successful to a great extent, the conventional load shedding schemes have certain disadvantages. The amount of a load step is, at times, large which causes excessive load to be shed. Most schemes do not have the flexibility to increase the number of load shedding steps, thereby introducing transients in the system. The most LS schemes proposed so far used voltage and frequency parameters, separately and also, the under-frequency and under-voltage relays are working in the power system without any coordination. The individual use of these indices may be also not reliable/effective, and may even lead to the over load shedding problems. Studying on the under-frequency load shedding is often done using the system frequency response models. The impact of voltage variation on the frequency deviation is not considered in these models. Furthermore, the UVLS methods that are proposed so far for adjusting the under-voltage relays, does not consider the frequency behavior. These two parameters (voltage and frequency) are not independent and the coordination between UFLS and UVLS schemes is therefore crucial. The dependency between voltage and frequency will affect LS performance. Economical considerations need to be considered before shedding the load since certain loads cannot be kept offline. Further, Load shedding is an emergency control operation and should be on a priority basis, which means shedding less important loads, while expensive industrial loads are still in service. Therefore, this study focuses on developing an algorithm that is more reliable and effective than the conventional schemes, that uses voltage and frequency parameters simultaneously for making load shedding decisions. This will help in power system planning, operation and control.
7.3
Relevance of Study/Justification
The increase in electrical power consumption is directly proportional to the increase in people population. As of 10th July, 2009, the world population was estimated by the United States Census Bureau at 6.77 billion. This figure is expected to reach about 9 billion by the year 2040. Many environmental and economic constraints prevent the construction of new or upgrading of the existing generation and transmission capacities. Additionally, generation reserves are minimal and often the reactive power is insufficient to satisfy the load demands. 27
Given this trend, power systems are expected to be more heavily loaded and moving closer and closer to stability limit and more susceptible to disturbances and outages.
Blackouts of power systems always have been a historical problem in interconnected power systems. However in recent years by improving monitoring and protection techniques, it is not possible to completely prevent of blackouts [1-3]. Sudden and large changes in generation capacity such as the outage of a generator can produce a sever imbalance between generation and load demand. This may lead to a rapid decline in frequency, because the system may not respond fast enough. If voltage and frequency get out from permissible range the system is in unstable condition. In this condition the system controller's operate and attempt to restore the voltage and frequency in the permissible range. If the disturbance is so large the controller's cant restore the voltage and frequency in the permissible range. In this condition the last solution to avoid the power system breakdown has been load shedding strategy.
Recent blackouts have brought our attention to the issues of voltage stability in the system. Voltage decline can be a result of a disturbance. Its main cause, however, is insufficient supply of reactive power. This has led researchers to focus on techniques to maintain voltage stability. The loss of a generator causes an unbalance between the generated power and the load demand. This affects the frequency and voltage. Load shedding schemes must consider both these parameters while shedding load. By shedding the correct amount of load from the appropriate buses, the voltage profile at certain buses can be improved [4].
While considering the amount of load to be shed and the step size, it is also important to take into account the reactive power requirements of each load. Quite often, disturbances such as a generator loss cause the voltage to decline. An effective way to restore voltage is to reduce the reactive power demand. Thus when loads absorbing a high amount of reactive power are first shed; the voltage profile can be improved.
7.4
Objectives
7.4.1 Main objective The main objective of the research is to investigate the applicability of a metaheuristic load shedding algorithm using both frequency and voltage parameters.
28
7.4.2 Specific objectives The specific objectives of this research include: a) Develop a robust optimal metaheuristic load shedding algorithm for UFLS and UVLS b) Perform optimal load shedding for UFLS and UVLS in a standard IEEE test bus system. c) Compare the performance of the algorithm with the existing schemes.
1.4
Scope of Study
The research will mainly investigate the most recent metaheuristic tools of Harmony search and cuckoo search and their applicability to power systems load shedding. The IEEE test bus systems will be used. 1.5 Importance of the study
The research will introduce a more effective tool in load shedding that can be utilized in optimal load shedding especially when multiple contingencies occur. The research will show how the dependency of voltage and frequency affect the performance of load shedding.
29
LITERATURE REVIEW
Different methods for load shedding and restoration have been developed by many researchers. Currently there are various under frequency and under voltage load shedding techniques used in the power industry worldwide.
2.1
Under voltage Load Shedding
Lopes et all [5] suggests a method which carries out load shedding in case of two conditions. One, where the load shedding occurs due to a post disturbance low voltage condition and secondly, where the load shedding results due to the inability of the system to achieve a stable operating condition during post disturbance. This method uses the load flow in order to decide the buses from which to shed load. The initial set of control actions are first carried out. These actions are capacitor switching, tap changing transformer and secondary voltage control.
Jianfeng et al [6] have developed a method with risk indices in order to decide which buses should be targeted for load shedding to maintain voltage stability. The buses with a high risk of voltage instability are considered first. This is estimated from the probability of a voltage collapse occurrence. The risk indices are the products of these probabilities and impact of voltage collapse.
Another method [7][31] dealing with the particle swarm approach for under voltage load shedding has been researched. The particle swarm Optimization concept is a group or cluster of particles in which each particle is known to have individual memory like an animal in its herd or flock. The flock is initiated with some initial velocity and the particles move in different directions to come up with the best solution. The best solution is shared with every particle of the group so that they can move from there on based on this new acquired knowledge. This same idea is used for under voltage load shedding to recognize the best possible load shedding scheme considering the system conditions and disturbance particular to that situation.
Ladhani and Rosehart [8] propose load modeling for an under voltage load shedding scheme. They also suggest offering economic incentives to customers for discontinuing the use of power during load control periods. This way the brunt of a sudden load shed is not borne by the customer alone. Also, systematic load control will lead to the stability of the system even when it is not faced with a disturbance.
30
Yorino et al [9] suggests a new planning method for planning the VAR allocation using the FACTS devices. Here, the total economic cost for a voltage collapse along with its corrective control and load shedding are taken into account to come up with the optimum VAR planning scheme. Thus, the objective function is to minimize the cost while keeping in mind the voltage stability of the system.
Mozino [10] discusses the currently existing under voltage load shedding schemes. They are divided into two categories; decentralized and centralized. The decentralized load shedding involves setting relays at buses with loads to be shed and tripping the respective relays. The centralized scheme is more advanced. The relays are installed at the key bus locations and the information regarding which relays are to be tripped is sent to these relays from a main control centre. Thus the required load is shed from appropriate buses. Many of these schemes are referred to as special protection or wide area schemes. The two categories mentioned above are widely used as under voltage load shedding relays. These relays require logic and have to perform efficiently and accurately. Also, these relays must avoid false operation. Thus to satisfy the above requirements digital relays are being used for under voltage load shedding.
Single Phase UVLS Logic measures voltages on every phase. This scheme distinguishes between voltage collapse and fault induced low voltages. The voltage collapse is a balanced phenomenon, hence results in a reduction of voltage on all the three phases. Except for a three phase fault all the other faults are unbalanced. The relays trips when it identifies a voltage collapse and blocks the relay for a fault induced low voltage. Unbalanced faults usually induce negative sequence voltages which are detected and used for blocking the relay. Positive sequence UVLS logic checks the positive sequence voltage with the set point value. Since the voltage collapse is balanced for all the three phases, the positive sequence voltage is equal to the three phase voltages. In case of a fault condition, the negative sequence voltage is utilized to block the relay.
A load shedding scheme against long term voltage instability is proposed by Van Cutsem et al [11]. It uses distributed controllers which are delegated a transmission voltage and a group of loads to be controlled. Each controller acts in a closed loop, shedding loads that vary in magnitude based on the evolution of its monitored voltage. Each controller acts on a set of 31
electrically close loads and monitors the voltage V of the closest transmission bus in that area. The controller is rule based where the rules are simple if-then statements.
2.2
Under frequency Load Shedding
Under frequency load shedding mainly sets up relays to detect frequency changes in the system. As soon as the frequency drops below a certain value a certain amount of load drops, if the frequency drops further, again a certain amount of load is dropped. This goes on for a couple of steps. The amount of load to be shed and the location of the load to be shed is predetermined. Terzia [12] talks about under frequency load shedding in two stages. During the first stage the frequency and rate of frequency changes of the system are estimated by non-recursive Newton-type algorithm. In the second algorithm, the magnitude of the disturbance is estimated using the simple generator swing equation.
In another approach Thalassinakis et al [13] obtained results from an autonomous power system on the Greek Islands of Crete. The method uses the Monte Carlo simulation approach for the settings of load shedding under frequency relays and selection of appropriate spinning reserve for an autonomous power system. The settings of the under frequency relays are based on the four parameters; the under frequency level, rate of change of frequency, the time delay and the amount of load to be shed. Three sets of system indices are defined. These sets are for the purpose of comparisons between load shedding strategies. A method was developed which simulated the behavior of a power system. The three aspects of the power systems that were developed in the simulation were Operation of the power system as performed by the control centre. Primary regulation of the generating units after the failure of a generating unit. Secondary regulation and utilization of the spinning reserves.
Three different cases of comparing the spinning reserves with the load mismatch are considered. One, when the spinning reserve is sufficient or greater. Thus the load can be restored immediately. Second, when the spinning reserve is slightly insufficient and the rapid generating units will require a certain amount of time to be started. Thus it will be 10-20 minutes before the load can be completely restored. Third, the spinning reserves are insufficient and there are not enough rapid generating units thus implying that the load will not be restored for a considerably long period of time.
32
Another method [14] triggers the under frequency relays based on a dynamically changing intelligent load shedding scheme. The main components of this scheme are the knowledge base, disturbance list and the ILS computation engine.
Fig. 1 Block Diagram of the ILS scheme
The generalized structure of the ILS scheme is shown in figure 1. The knowledge base is the most important block. It is connected to the computation engine which sends trip signals to relays. The network models can be accessed by the knowledge base while monitoring the system. The knowledge base is trained and its output consists of system dynamic scenarios and frequency responses during disturbances. This trained knowledge base also monitors the system continuously for all operating conditions. The disturbance list consists of pre-specified system disturbances. Based on the inputs for the system and the continuous system updates, the knowledge base notifies the ILS engine to update its load shedding list. Thus it ensures that the load shed is always minimum and optimum.
Wee-Jen Lee [15] discuss about another intelligent load shedding based on microcomputers. The unique feature about this scheme is the built in frequency setting and the time delay setting. The frequency setting in the relay counters system re collapse situation. (Consider a generator loss which triggers a load shedding step. This causes the frequency of the system to recover. During this recovery period if another generator trips it results in a system re collapse). Typical frequency relays will not trip until the second generator loss causes sufficient frequency decay. The ILS system automatically adjusts the frequency settings such that load is shed immediately without delay. The time delay settings cause the load scheme to initiate during situations when a disturbance causes the frequency to drop and hold at a value less than the rated. The number of load shedding steps can be increased without a limit. The advantage of having large number of 33
load shedding steps is that it prevents large amount of transients. It also prevents over shedding. Denis Lee Hau Aik, [16] suggests a method using the System Frequency Response SFR and the Under Frequency Load Shedding UFLS together to get a closed form expression of the system frequency such that the UFLS effect can be included in it. On doing this, the system and UFLS performance indicators can be calculated. Thus these indicators can be used efficiently in any further optimization techniques of SFR UFLS model. One such method has been discussed using the regression tree by Chang et al [17]. The regression tree is utilized to interpolate between recorded data to give an estimate of the frequency decline after a generator outage. It is a non parametric method which can select the system parameters and their relations which are most relevant to the load imbalance (due to generator outage) and the frequency decline. The case considered here is only a generator outage but this method can be applied to other forms of disturbances as well.
A Kalman filtering-based technique by A.A. Girgis et al [18] estimates frequency and its rate of change which is beneficial for load shedding. The noisy voltage measurements are used to estimate the frequency and its rate of change. A three-state extended Kalman filter in series with a linear Kalman filter is used in a two stage load shedding algorithm. The output of the three stage Kalman filter acts as the input to the linear Kalman filter. It is the second filter which identifies linear components of the frequency and its rate of change. The amount of load to be shed is calculated using the linear component of the estimated frequency deviation.
Another method uses Kalman filtering [19] to estimate the frequency and its rate of change from voltage waveforms. The buses are ranked based on their rate of change of voltage (dV/dt) values. The disturbance magnitude is calculated from the swing equation. The rate of change of frequency required for this equation is calculated using the Kalman filter. Once the total amount of load to be shed is estimated then the load to be shed from each bus is determined based on the PV analyses.
An optimization technique for load shedding [20] with distributed generation was developed. This technique converts differential equation into algebraic ones using the discretization method. Two cases are considered here; one with the distributed generation switched on to the system as a static model and the other case without the distributed generation on the grid. Both cases resulted in successful shedding of appropriate quantity of load. 34
Li Zhang suggests a method [21] which designs under frequency relays using both the frequency and the rate of change of frequency (df/dt). The scheme has been designed for a 50 Hz Northeast China power system. Traditional schemes required only the frequency decay information. Here the rate of change of frequency is used as auxiliary information. The plots for the rate of change of frequency are oscillatory in nature. Hence a new scheme is devised in this paper which considers the integration of the rate of change of frequency (df/dt) to indicate the frequency drop. By integrating one is effectively measuring the area between two frequencies, fi-1 and fi. The schemes is made up of five load shedding steps for a 50 Hz system. These steps are from 50 to 49.2 Hz, 49.2 to 49 Hz, 49 to 48.8Hz, 48.8 to 48.6 Hz, 48.6 to 48.4 Hz. The amount of load to be shed in each step is decided by integrating the df/dt value in each step. The simulation results when compared with the old scheme with just the frequency decay show a definite improvement in system frequency due to the inclusion of rate of change of frequency (df/dt) in the new scheme.
The main idea in the paper proposed by Xiong et al [22] is the inclusion of on line load frequency regulation factors. Loads with smaller frequency regulation factors are shed first, followed by the ones with larger frequency regulation factors. The active power and load frequency relation is established in the form of the following equation.
Where, fN is the nominal frequency. PLN is the rated active power and ai (i=1,2n) is the percentage of the total load associated with the i-th term of the frequency. The per unit form of the above equation is differentiated to get the change in load power as frequency changes (dPL/df) which is the KL factor or regulation factor. The higher order terms are neglected.
35
Another scheme considering the rate of change of frequency is the adaptive load shedding algorithm in the paper by Seyedi et al [23]. Here the shedding is adapted as per the intensity of the disturbance. This intensity is determined based on the rate of change of frequency. Thus the main points observed while designing the scheme is that the speed of load shedding is increased if the rate of change of frequency (df/dt) values are high. Also, the number of load shedding steps and the amount of load to be shed in each step is increased if there is an increase in the rate of change of frequency (df/dt) values. The new method was tested on the HV network of the Khorasan province in Iran. The proposed method definitely showed improvements as compared to the conventional scheme.
Neural networks are proposed [24] to be used for an under frequency load shedding scheme. This intends to replace the conventional slow acting dynamic simulators by quick and efficient neural network engines. The general procedure is to identify the inputs for the neural networks, generations of data sets, designing NN and the evaluating the performance of neural nets. The variables used as inputs are the actual real power generation, available real power, actual load generation level prior to a disturbance, amount of the actual load being shed and the percentage of the exponential load to be shed.
A SCADA based scheme has been proposed by Parniani et al [25]. The rate of change of frequency is useful in identifying the overload when a disturbance occurs and hence is helpful to estimate the amount of load to be shed. The SCADA based scheme overcomes the shortcomings of the previous adaptive UFLS scheme. The mean system frequency is defined as follows,
Where fi is the frequency of the generators from 1 to n and H is their respective system inertia. Adding the df/dt equation every generator the post disturbance equation obtained for SCADA is
36
where is the disturbance magnitude in per unit. Now another variable is defined. If a disturbance occurring at the weakest generator is less than this value then absolute frequency of that generator is within the permitted limits. For a situation where the disturbance magnitude, is less than no load shedding is required. The maximum load shedding magnitude is equal to the difference between the disturbance magnitude and Pthr - Pthr. The load to be shed is distributed inversely proportional to the generator inertia to make the load shedding most effective. The equation (4) represents this distribution.
Based on this equation the layers of the load shedding scheme are designed. Both the steps shed one third of the remaining load. These are in steps. They are presented in a table 1 with the first step being at 59.3 Hz.
TABLE 1: SCADA Based Load Shedding Formula
An adaptive load shedding scheme which includes a self healing strategy is presented by Vittal et al [26]. The proposed scheme is tested on a 179 bus 20 generator test system. This self healing strategy comes into play when the system vulnerability is detected. The system then divides into self sustaining islands. After this islanding, load shedding based on the rate
37
of change of frequency is applied to the system. Due to this division, it becomes easier to restore load. A Reinforcement Learning scheme is discussed in the paper. The first is the controlled islanding which is done using the two-time scale method. It deals with the structural characteristics of the power systems and determines the interactions of the generators and their strong or weak coupling. The Dynamic Reduction Program 5.0 (DYNRED) is the software in which simulations are run to implement this technique. Through this software coherent group of generators can be obtained on the power system. Islanding causes two types of islands to be formed, the generation rich islands and the load rich islands. The load rich islands may have a further decline of frequency. This may result in the generator protection to trip the generators thus further declining the islands frequency. Thus a two layer load shedding strategy is employed for the load rich island. The first layer is based on the frequency decline approach. The second layer considers the rate of change of frequency. Due to the longer time delays and lower frequency thresholds for a frequency based scheme inadvertent load shedding is avoided. When the system disturbance is large and exceeds the signal threshold, the second layer comes into play. It sends a signal to discontinue the first layer of operation and continues with the load shedding based on rate of change of frequency. This layer will shed more load at the initial steps to prevent cascading effects. The magnitude of the disturbance is found based on the formula
If we sum up all the equations for i=1 to n then the final equation obtained is
Where, m0 is defined as df/dt which is the average rate of frequency decline. Rearranging the above equation we get a new equation which relates PL to m0 .
38
Since Hi is constant, the magnitude of m0 can be directly proportional to the rate of frequency decline. Hence the rate of change of frequency df/dt ) can be a measure of the disturbance. Once the disturbance threshold value, PL , for the second layer of load shedding is decided, the m0 value is calculated. The mi at each bus is calculated and compared with m0 . If mi m0
then the second layer is activated, otherwise the conventional load shedding scheme is used. This new shedding scheme increases the stability of the system by shedding fewer loads as compared to the conventional scheme.
Application of Neural Network in load shedding and some Predictable functioning of load shedding methods has been proposed [28 -30][32]. In this method the identification of the variables like inputs and outputs is an important step for a successful application of this technique. Sometimes a pre-processing stage is needed to choose the most significant variables to be used as inputs of a NN. Some of the meaningful variables that have been used as inputs of the NN Active real power generation Active load generation Amount of active load being shed Percentage of exponential type loads being shed Damping factor Power factor
These variables provide the NN with valuable information, such that it can make the required assessment with respect to how much the generation load disproportion has been corrected and the influence each load type has on the resulting frequency response. 2.3 Under voltage under frequency Load Shedding
A load shedding scheme that incorporates, the frequency and the bus voltages, for deciding the instant, the amount and the location of the load to be shed is proposed [27]. The scheme developed consists of a stepwise approach. This has been represented in the form of an algorithm. The first step of the load shedding procedure is the measurement and calculation of the rate of change of frequency. Depending on the relay, the frequency measurements or the rate of change of frequency are recorded in the system. The total load mismatch between the generated power and load power is determined. For a single machine, the swing equation [1] is given by 39
where, f0 is the nominal frequency of the system and Pdiff is the difference in the generated power and the load power. In the above equation is replaced by f since = 2f . Thus a relation between the frequency and the power mismatch is obtained. This relation establishes the estimated magnitude of the disturbance. The inertia constant in the above equation is the kinetic energy Wk over the system base MVA. The inertia constants of all the machines in the system are on the base MVA. In a large power system where there are many generators which maybe geographically far away from each other,
Also, the equivalent mechanical and electrical powers are given as;
Pm = individual mechanical shaft power of each machine for all the machines in the system Pe = individual electrical power of each machine for all the machines in the system
Once the magnitude of the disturbance is determined using the above equivalent swing equation, the location and the amount of load to be shed from each bus has to decided. In order to do this, the buses are ranked according to the dV/dt values at the point of detection of frequency decline. The bus with the largest dV/dt is listed at the top of the list and then so on in the decreasing order. Once the order is decided, the next step is to decide the amount of load to be shed at each bus. This is decided based on the voltage sensitivity at each bus. Thus the bus with voltage sensitivity very close to the instability limit will have a maximum load shed based on the reciprocal of its sensitivity as a fraction of the sum of the reciprocals of all the load bus sensitivities.
40
The load shedding scheme has been tested on the IEEE 39 bus and IEEE 145 bus test systems. Three studies based on contingency were carried out with each contingency being considered at a time: Case study 1: Loss of a generator for the IEEE 39 bus system. Case study 2: Loss of a generator for the IEEE 145 bus system. Case study 3: Loss of a transmission line IEEE 39 bus system
The scheme is simple and does not involve complicated calculations. It proved to be successful in restoring the frequency within its pre-defined limits. It has also improved the voltage profile at certain buses which had critically low voltage before load shedding was applied. However the algorithm did not address some aspects and therefore requires fine tuning. Economical considerations need to be considered before shedding the load since certain loads cannot be kept offline. The study and testing of the scheme in a multiple contingency scenario like loss of a generator along with a loss of transmission line need to be considered as this would create a critical situation.
Thus the various conventional schemes, under frequency schemes and under voltage load shedding schemes have been discussed above. These give an insight about the technological advancement achieved in this area. The proposed study intends to investigate the applicability of harmony search algorithm as an optimization tool and how it can be applied to address the disadvantages faced by the conventional schemes present in the industry.
41
3 3.1
METHODOLOGY Metaheuristics algorithm
Metaheuristic algorithms are higher-level heuristic algorithms. Here, meta-means higherlevel or beyond, so metaheuristic means literally to find the solution using higher-level techniques, though certain trial-and-error processes are still used. Broadly speaking, metaheuristics are considered as higher-level techniques or strategies which intend to combine lower-level techniques and tactics for exploration and exploitation of the huge space for parameter search.
There are two important components in modern metaheuristics, and they are: intensification and diversification. For an algorithm to be efficient and effective, it must be able to generate a diverse range of solutions including the potentially optimal solutions so as to explore the whole search space effectively, while it intensifies its search around the neibourhood of an optimal or nearly optimal solution. In order to do so, every part of the search space must be accessible though not necessarily visited during the search. Diversification is often in the form of randomization with a random component attached to a deterministic component in order to explore the search space effectively and efficiently, while intensification is the exploitation of past solutions so as to select the potentially good solutions via elitism or use of memory or both [33-35]. If the intensification is too strong, only a fraction of local space might be visited, and there is a risk of being trapped in a local optimum, as it is often the case for the gradient-based search such as the classic Newton-Raphson method. If the diversification is too strong, the algorithm will converge too slowly with solutions jumping around some potentially optimal solutions. In this study good balance of these two important components will be maintained. Another important feature of modern metaheuristics is that an algorithm is either trajectorybased or population-based. It is difficult to decide which type of method is more efficient as both types work almost equally successfully under appropriate conditions. In this study the focus will be on population based algorithm.
3.2
Harmony Search Algorithm
In the HS algorithm, diversification is essentially controlled by the pitch adjustment and randomization -- here there are two subcomponents for diversification, which might be an important factor for the high efficiency of the HS method. The first subcomponent of composing new music, or generating new solutions, via randomization would be at least at
42
the same level of efficiency as other algorithms by randomization. However, an additional subcomponent for HS diversification is the pitch adjustment characterized by rpa. Pitch adjusting is carried out by adjusting the pitch in the given bandwidth by a small random amount relative to the existing pitch or solution from the harmony memory. Essentially, pitch adjusting is a refinement process of local solutions. Both memory consideration and pitch adjusting ensure that the good local solutions are retained while the randomization and harmony memory considering will explore the global search space effectively. The subtlety of this is that it is a controlled diversification around the good solutions (good harmonics and pitches), and it almost acts like an intensification factor as well. The randomization explores the search space more efficiently and effectively; while the pitch adjustment ensures that the newly generated solutions are good enough, or not too far away from existing good solutions. The intensification is mainly represented in the HS algorithm by the harmony memory accepting rate raccept. A high harmony acceptance rate means the good solutions from the history/memory are more likely to be selected or inherited. This is equivalent to a certain degree of elitism. Obviously, if the acceptance rate is too low, the solutions will converge more slowly. Furthermore, the HS algorithm is a population-based metaheuristic, this means that multiple harmonics groups can be used in parallel. Proper parallelism usually leads to better implantation with higher efficiency. The good combination of parallelism with elitism as well as a fine balance of intensification and diversification is the key to the success of the HS algorithm, and in fact, to the success of any metaheuristic algorithms. These advantages make it very versatile to combine HS with other metaheuristic through hybridization .This research will focus on hybridization of Harmony search algorithm with Cuckoo search optimization.
3.3
Cuckoo Search Optimization
Cuckoos are brood parasites that lay their eggs in the nests of other birds (such as crows) who serve as hosts to hatch their eggs. To elucidate, let there be n parasites and equally many (although not necessarily) hosts. Each parasite individual would be represented by a point (x in m-dimensional space) and similarly each host individual would be represented by a point (y in m-dimensions). These points may be randomly generated and would lie in the domain of the function to be optimized. Each cuckoo would take a Lvy flight and if its post-flight fitness is better than its pre-flight fitness, it would randomly choose a host nest that has not as yet been invaded by another cuckoo and the quality of the host eggs are inferior to the cuckoo egg. If this condition is not met, it would not lay any egg in the host nest. The egg of a 43
successful parasite may, however, be detected (with probability p) by the host and be destroyed. If not detected, however, it would be hatched in the host nest and eventually join the cuckoo population. Only the best n cuckoos, however, would enter into the next generation. To implement this search scheme, Yang & Deb[37] formulated the following idealized rules: (a) Each cuckoo lays a single egg into a randomly chosen host nest from among n nests; (b) The nests with better quality eggs (implying better fitness value of the function concerned), if not detected, would be hatched to grow into the cuckoo chicks, who would join the next generation; (c) The number of available host nests is fixed. The host can detect the alien egg with a probability [0, 1] and, if detected, it will either abandon the nest and build a new nest elsewhere or destroy the egg; (d) When generating new solutions xi(t+1) from the old one xi(t), Levy flight is performed with parameter 1< < 3 and thus
The Lvy flight is a type of random walk which has a power law step length distribution with a heavy tail.
44
THESIS DEVELOPMENT SCHEDULE
YEAR 1
May-10 Mar-14 Aug-13 Aug-14 Nov-13 Apr-14 Dec-13 Feb-14 Jun-14 Sep-13 Oct-13 Jan-14 Sep-14 Oct-14 Jul-13 Jul-14
YEAR 2
May-15 Mar-15 Aug-15 Nov-14 Apr-15 Dec-14 Feb-15 Jun-15 Jan-15 Sep-15 Oct-15 Jul-15
YEAR 3
May-16 Mar-16 Nov-15 Apr-16 Dec-15 Feb-16 Jun-16 Jan-16
ACTIVITY
Proposal Editing and Presentation Literature Review: Harmony search algorithm study and coding Load shedding study using harmony search algorithm Paper Presentation Hybridization of harmony search with Cuckoo search algorithm Load shedding study using the hybrid algorithm Paper Presentation Thesis Writing & Presentation Thesis Examination Thesis Corrections and editing Thesis Defence
45
BUDGET
ITEM Stationery, Printing, Photocopying for entire research period Computer Soft ware ( C++, Matlab and associated tool boxes) Computer Hardware: Printer, Laptop Access to journals: IEEE, Actpress Transport Books Conferences and Paper presentation Thesis printing and binding Total
COST (KSH)
40,000
2 3 4 5 6 7 8
80,000 80,000 50,000 50,000 40,000 360,000 20,000 720,000
46
6 [1]
REFERENCES D. Craciun, S. Ichim, and Y. Besanger, A new soft load shedding: Power system stability with contribution from consumers, IEEE/PTC, pp. 1-6, June/July 2009. D.Andersson, P.Elmersson, A.Junti, Z.Gajie, D.Karlsson, and L.Fabiano, Intelligent load shedding to counteract power system instability, IEEE/PES, pp. 570-574, 2004. G. Shahgholian, M. E. Salary Effect of Load Shedding Strategy on Interconnected Power Systems Stability When a Blackout Occurs, International Journal of Computer and Electrical Engineering, Vol. 4, No. 2, pp 212-217, April 2012 B. Suneetha, S. Hemachandra, B. Gowri Prasad and A. Jayanth, Application of Neural Network in Load Shedding and Some Predictable Functioning of Load Shedding Methods, International Journal of Electronics, Communication & Instrumentation ngineering Research and Development (IJECIERD), Vol. 3, Issue 2, pp 45-50, Jun 2013
[2]
[3]
[4]
[5]
B. Isaias Lima Lopes, A.C. Zambroni de Sow, An Approach for UnderVoltage Load Shedding, IEEE Bologna PowerTech Conference.,2003
[6]
Dai Jianfeng, Zhou Shuangxi, Lu Zongxiang, A New Risk Indices Based Under Voltage Load Shedding Scheme., Transmission and Distribution Conference and Exhibition: Asia and Pacific, 2005 pg 1-5
[7]
T. Amraee, B. Mozafari, A.M. Ranjbar, An Improved Model for Optimal Under Voltage Load Shedding: Particle Swarm Approach. Power India, New Delhi Conference, IEEE, 2006
[8]
Shamir S. Ladhani, William Rosehart, Criteria for Load Control when Considering Static Stability Limits. Power Engineering Society General meeting, 2005, IEEE,vol 2 pg 1558 1562
[9]
Naoto Yorino, E. E. El-Araby, Hiroshi Sasaki, Shigemi Harada, A New Formulation for FACTS Allocation for Security Enhancement Against Voltage Collapse.Power Systems, IEEE transactions, 2003, vol 18,issue 1,pg 3-10
[10]
Charles J. Mozina (Beckwith Electric Co., Inc.), Undervoltage Load Shedding., Power Systems Conference: Advanced Metering, Protection, Control, Communication, and Distributed Resources, 2007. Pg 39-54
[11]
Mark Adamiak, William Premerlani, Dr. Bogdan Kasztenny, Synchrophasors: Definition, Measurement, and Application.
[12]
[21] Vladimir V. Terzija, (2006) Adaptive Underfrequency Load Shedding Based on 47
the Magnitude of the Disturbance Estimation, IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 21, NO. 3. [13] Emmanuel J. Thalassinakis, Member, IEEE, and Evangelos N. Dialynas,(2004) A Monte-Carlo Simulation Method for Setting the Underfrequency Load Shedding Relays and Selecting the Spinning Reserve Policy in Autonomous Power Systems, IEEE TRANSACTIONS ON POWER SYSTEMS, 2004,VOL. 19,NO. 4. [14] Farrokh Shokooh, Shervin Shokooh, Jacques Tastet, Hugo Castro, Tanuj Khandelwal, Gary Donner, An Intelligent Load Shedding System Application in a Large Industrial Facility, Industry Applications Conference, 2005. Fourtieth IAS Annual Meeting. Conference Record of the 2005 (Volume:1 ) pg 417-425 [15] Wen-Jen Lee, Microcomputers based intelligent load shedding. Power Delivery, IEEE Transactions on (Volume:4 , Issue: 4 ) 2002, pg 2018 - 2024 [16] Denis Lee Hau Aik, Member, A General-Order System Frequency Response Model Incorporating Load Shedding: Analytic Modeling and Applications, IEEE
TRANSACTIONS ON POWER SYSTEMS, 2006, VOL. 21,NO.2. [17] Rung-Fang Chang, Chan-Nan Lu, Tsun-Yu Hsiao, Prediction of Frequency Response After Generator Outage Using Regression Tree, IEEE TRANSACTIONS ON POWER SYSTEMS, 2005, VOL. 20, NO. 4. [18] Adly. A. Girgis, William Peterson, Adaptive estimation of frequency deviation and its rate of change for load shedding, Power Delivery, IEEE Transactions on (Volume:5 , Issue: 2 ),2002 [19] Adly.A.Girgis, Shruti Mathure, Dissertation on Load shedding using voltage, frequency and their rate of change. [20] Adly.A.Girgis, Ding Xu, Optimal Load shedding strategies in power systems with distributed generation. Power Engineering Society Winter Meeting, 2001. IEEE Volume:2 pg 788-793 [21] Li Zhang, Jin Zhong, UFLS Design by using f and integrating df/dt., Power Systems Conference and Exposition, 2006. Pg 1840-1844 [22] Xiaofu Xiong, Wenyuan Li, A New under frequency load shedding scheme considering load frequency characteristics. Power System Technology, 2006. PowerCon 2006. Pg 1-4 [23] H. Seyedi, M. Sanaye-Pasand, M.R. Dadashzadeh, Design and Simulation of an adaptive
Load Shedding Algorithm using a real network. Power India Conference, IEEE, 2006 [24] Matthew A. Mitchell, J.A. Peas Lopes, J.N. Fidalgo James D. McCalley,
48
Using a Neural Network to Predict the Dynamic Frequency Response of a Power System to an Under-Frequency Load Shedding Scenario. Power Engineering Society Summer Meeting, 2000. IEEE (Volume:1 ) pg 346 - 351 [25] M. Parniani, A. Nasri, SCADA based under frequency load shedding integrated with rate of frequency decline. Power Engineering Society General Meeting, 2006. IEEE [26] Juhwan Jung, Chen-Ching Liu, Steven Tanimoto, Vijay Vittal, Adaptation in Load Shedding under vulnerable operating conditions. Power Systems, IEEE Transactions on (Volume:17 , Issue: 4 ) pg 1199-1205 [28] B. Suneetha, S. Hemachandra, B. Gowri Prasad & A. Jayanth, Application Of Neural Network in Load Shedding and some Predictable Functioning of Load Shedding Methods, International Journal of Electronics, Communication & Instrumentation Engineering Research and Development (IJECIERD) Vol. 3, Issue 2, Jun 2013, pg 45-50 [29] E. J. Thalassinakis, E. N. Dialynas, D. Agoris, Method Combining Ann's and Monte Carlo Simulation for the Selection of the Load Shedding Protection Strategies in Autonomous Power System, IEEE Trans. on Power System, Vol. 21, No.4, pp. 1574-82, 2006. [30] C. T. Hsu, M. S. Kang and C. S. Chen, Design of Adaptive Load Shedding by Artificial Neural Networks, IEE Generation, Transmission, Distribution, Vol. 152, No. 3, pp. 415421, 2005. [31] T. Amraee, B. Mozafari, A. M. Ranjbar, An Improved Model for Optimal Under Voltage Load Shedding, Particle Swarm Approach, IEEE Power Conf. India, April 2006. [32] Gh. Isazadeh, R. Hooshmand and A. Khodabakhshian, Design Of An Adaptive Dynamic Load Shedding Algorithm Using Neural Network in the Steelmaking Cogeneration Facility, IJST, Transactions of Electrical Engineering, Vol. 36, No. E1, Pp 67-82,2012 [33] [34] [36] Yang XS (2008) Nature-inspired Metaheuristic Algorithms. Luniver Press Glover F and Laguna M (1997) Tabu Search. Kluwer Academic Publishers Yang, X.S. & Deb, S. (2009) Cuckoo search via Lvy flights, Proc. World Congress on Nature & Biologically Inspired Computing (NaBIC 2009, India), IEEE Publications, 210 214.
49

Research Proposal

Caricato da

Informazioni sul documento

Descrizione originale:

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Research Proposal

Caricato da

Copyright:

Formati disponibili

JOMO KENYATTA UNIVERSITY OF AGRICULTURE AND TECHNOLOGY

PhD Research Proposal

INTRODUCTION Relevance of Study/Justification

Under frequency Load Shedding

Fig. 1 Block Diagram of the ILS scheme

PL is the disturbance magnitude in per unit. Now another variable

magnitude is equal to the difference between the disturbance magnitude and

TABLE 1: SCADA Based Load Shedding Formula

Emmanuel J. Thalassinakis, Member, IEEE, and Evangelos N. Dialynas,(2004)

A METAHEURISTIC LOAD SHEDDING ALGORITHM USING VOLTAGE AND FREQUENCY PARAMETERS

J.K.U.A.T July, 2013

DR CHRISTPHER MAINA MURIITHI SIGN_______________DATE: ________

1.4.1 1.4.2 1.4 1.5 2

Scope of Study ........................................................................................................... 29 Importance of the study ............................................................................................. 29

Under voltage Load Shedding

Under frequency Load Shedding

Fig. 1 Block Diagram of the ILS scheme

TABLE 1: SCADA Based Load Shedding Formula

METHODOLOGY Metaheuristics algorithm

Harmony Search Algorithm

Cuckoo Search Optimization

THESIS DEVELOPMENT SCHEDULE

80,000 80,000 50,000 50,000 40,000 360,000 20,000 720,000

[21] Vladimir V. Terzija, (2006) Adaptive Underfrequency Load Shedding Based on 47

Potrebbero piacerti anche

DR CHRISTPHER MAINA MURIITHI SIGN_______DATE: