Sei sulla pagina 1di 16

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/328751289

A Sliding Mode Control Approach for Gas Turbine Power Generators

Article  in  IEEE Transactions on Energy Conversion · November 2018


DOI: 10.1109/TEC.2018.2879688

CITATIONS READS

7 129

6 authors, including:

Andrea Bonfiglio Damiano Lanzarotto


Università degli Studi di Genova Università degli Studi di Genova
31 PUBLICATIONS   251 CITATIONS    11 PUBLICATIONS   54 CITATIONS   

SEE PROFILE SEE PROFILE

Alessandro Palmieri R. Procopio


Università degli Studi di Genova Università degli Studi di Genova
6 PUBLICATIONS   17 CITATIONS    128 PUBLICATIONS   1,315 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

High Frequency Fields analysis for TLC applications View project

Innovative Control Techniques for Primary Regulation in Islanded Microgrids View project

All content following this page was uploaded by Alessandro Palmieri on 27 November 2018.

The user has requested enhancement of the downloaded file.


1

A Sliding Mode Control Approach for Gas


Turbine Power Generators
Andrea Bonfiglio, Stefano Cacciacarne, Marco Invernizzi, Damiano Lanzarotto, Alessandro Palmieri,
Renato Procopio

Abstract—Gas Turbines are raising nowadays growing interest in the power generation industry. Being able to
regulate their power production faster than any other form of conventional source, they are extremely valuable in
compensating stochastic power fluctuations deriving from renewable energy sources in large scale electrical grids,
preserving grid stability, etc. In this context, gas turbines dynamical performance plays a key role as it needs to be
enhanced compatibly with preserving the machine lifetime. Nonlinear model based control systems can effectively
deal with this problem, exploiting the system structure to obtain optimal dynamical performance. Therefore, this paper
aims at developing a Sliding Mode model based control for gas turbines and at comparing the results obtained by its
application on a heavy duty Ansaldo Gas Turbine model (AE94.3A) with the ones obtained thanks to the traditional
controls employed in these applications.

I. INTRODUCTION
Energy scenario has experienced an increasing and noticeable change in recent years, shifting from a power generation
situation characterized almost by a totality of dispatchable power plants, to a situation dominated by an extensive
presence of renewable energy sources (RES) [1] [2]. Their stochastic and non-dispatchable production forced traditional
power plants to reach new levels of reliability, efficiency but most of all flexibility in terms of power generation
regulation. Gas turbines (GT) play a fundamental role in this scenario due to their ability to regulate power production
faster than any other form of conventional energy source [3] [4]. What makes interesting these components is their
suitability both for small and large size applications (i.e. aviation, ship applications, microgrids [5], and traditional
power plant applications). Gas turbine dynamic behaviors are sufficiently well described by nonlinear mathematical
models; because of their complexity, control systems used today in this sector are based on linear approximation of
these models, i.e. conventional PIDs [6]. These controllers are on the one hand simple to design, but on the other they
restrict gas turbine dynamical performance [6]. In this new environment, where more and more rapid power regulation
is required, together with preserving machine lifetime, new control techniques acquire a growing importance. For
these reasons, many authors recently focused on the study of new control systems built on innovative model based
control theories. Several works exploited the well-known gas turbine model presented by Rowen in [7], however this
model is linear and too simple to be used to design a high-performance model based controller. The same problem
arises in [8] where a gas turbine linear model is used to design a neuro –fuzzy control system. In order to cope with
gas turbine actual nonlinearities in [9] a “One-Step-Ahead” adaptive controller, able to self-tune its parameters
depending on the current work point, is designed, in [10] a H∞ robust control is applied, while in [11] the model
predictive control (MPC) technique is used. However, all these research are based on different approximations: the
inlet guide vanes (IGV) position rate limiter is not considered, the dynamic associated with the Exhaust Gas
Temperature (EGT) measurement is neglected and the technical impossibility of the Combustion Chamber
Temperature (CCT) measurement is not mentioned. In particular, the second approximation cancels the typical delay
affecting the EGT data acquisition; this fact combined with the absence of a CCT measurement makes this variable
critical for the machine lifetime [12] and for the generation of unstable flame phenomena [13] such as humming [14].
As a matter of fact, the slow dynamic of the sensor filters rapid temperature variations, making extremely difficult for
control systems to handle properly the CCT dynamic during fast load increase; resulting for instance in a reduction of
the machine lifetime. In addition, besides the EGT control, all these works aim to regulate the machine speed, which
is a valid approach only if the generator set supplies its own dedicated load. On the contrary, gas turbine control
systems in real applications on interconnected grid receive the power set point from the primary regulator [15],
therefore, besides the EG temperature control, the aim is not the machine speed regulation but the generated power
instead. In this perspective the Feedback Linearization (FBL) theory [16] was applied on a nonlinear Vth order gas
2

turbine model in [6] and compared with the conventional PID control. The approximations discussed above were
removed and the FBL controller showed increased performance although CCT spikes during rapid load increase were
not eliminated. However, the FBL controller, not being a robust control system, cannot handle the system in the
functioning area with IGV saturated, with the CCT roughly estimated and in the presence of other model uncertainties.
This is the reason why the CCT was assumed to be exactly measurable and the behavior of the gas turbine with IGV
saturated was not investigated. In particular, the IGV saturation problem is given by the fact that gas turbines can only
regulate the air flowing into the compressor from 100% to 60% of their maximum value. This implies that when the
IGV reach their maximum or minimum opening position, the air flow cannot be regulated, thus only one output can
be controlled. Despite the system change of behavior, the control must be kept on the power channel in these situations,
while the EGT value is uniquely given by the current work point. In order to overcome these limitations a robust
control system is needed. Many authors focused their attention on the sliding mode (SM) approach which recently has
been extensively applied on power systems, to name a few: in [17] the SM theory is applied to design a fuel cell
controller, in [18] a SM control system is realized to control active and reactive power for wind turbines, in [19] and in
[20] the SM approach is exploited to design robust position controllers for induction and DC motors and the list extends
further.
On balance, this paper aims to design a sliding mode robust controller according to the theory proposed in [16],
based on the Vth order gas turbine model mentioned above. In particular, the goal is to propose a SM control system
able to totally eliminate CCT spikes during rapid load increase, keep good performance in the functioning area with
IGV saturated, avoid the necessity of an exact CCT measurement and be robust to parametric and dynamic
uncertainties (i.e. model approximations).

II. GAS TURBINE MODEL AND STATE TRANSFORMATION


A. Gas Turbine model used for the control system design
The GT model used for the controller design is presented in detail in [6] together with the numerical value of its
parameters. The main assumptions on which this model is constructed are the following:
 The system is composed by three macro components: compressor, combustion chamber and turbine.
 A unique control volume is assumed for the combustion chamber, supposing that two state variables (temperature
and pressure), describe the thermodynamic characteristics of the fluid.
 Combustion chamber gas are considered to be ideal.
 Air and fuel flow actuator dynamics are supposed to be linear and of the 1st order.
In order to present the main steps leading to the SM controller design only the essential passages are presented here.
The system can be written in the form:
 x  f  x   g  x  u
 (1)
 y  h  x 
Thus the state, input and output vectors are respectively defined as follows:
x   x1 , x2 , x3 , x4 , x5    pcc , Tcc , m
 a ,m
 f ,Texm 
T T
(2)

u  u1 , u2    m
 *a , m
 *f 
T T
(3)

y   y1 , y2    Pgt , Texm 
T T
(4)
And:
f  x    f1  x  , f 2  x  , f 3  x  , f 4  x  , f 5  x  
T
(5)
g  x    g1  x  g 2  x  (6)

h  x    h1  x  , h2  x 
T
(7)
In particular:
3

  Ra

f1  x  
Re  c paTamb
x3   x1  c pa    1  
 
Vcc  c pe  Re   c  p 
c
   amb  
  (8)
TccN x1 
 x4 H f  c pe x2 M a0 
pccN x2 
  Ra

f2  x  
x2 Re  Tamb   x1  c pa 
  Re  cPe  x2  cPa  p 
x1Vcc  c pe  Re   c   amb 
   (9)
T x1 
c  1 x3   Re  cPe  x2  H f  x4  Re x2 M a0 ccN 
pccN x2 
1
f3  x    x3 (10)
 igv
1
f4  x    x4 (11)
 at
  Re
 
1   pamb  pe  1  1  x 
c
f5  x    x2t 5 (12)
 tc   x  t 
  1   

While the G-matrix vector components are:
T
 1 
g1  x   0 , 0 ,  , 0 , 0  (13)
  igv 
T
 1 
g 2  x   0 , 0 , 0 ,  , 0 (14)
  at 
And the output variables:
 Re

  p  c pe

h1  x   t kx1 x2 1   amb

 x1  
  
(15)
  Re
 
   pamb  c pe 1  
 x3 c pa  x2t      1  Tamb 
  x   
 
1 t
 
h2  x   x5 (16)

B. State transformation
In order to proceed with the SM control design the system has to be written in the companion form as shown in [16].
Thus, a coordinate change is needed via input/output FBL. However, this transformation leads to an unstable internal
dynamics [6], thus it implies that this representation of the system is not suitable to design this kind of controller.
However, this fact can be avoided by applying the FBL procedure to the same system replacing (15) with the following
approximated expression:
h1  x    P x1 x2 (17)
Hence, the FBL transformation develops as follows:
y1  L f h1  Lg h1  u1  Lg h1  u2
1 2
(18)
In which:
h1 h h h h
L f h1  f1  1 f 2  1 f3  1 f 4  1 f 5 (19)
x1 x2 x3 x4 x5
Thus, omitting some calculations:
4

L f h1   P  x2 f1  x   x1 f 2  x   (20)
And:
Lg1 h1  0 (21)
Lg2 h1  0 (22)
Both (21) and (22) equal to zero, it is then necessary to proceed with the derivative process until at least one input
appears:
y1  L2f h1  Lg L f h1  u1  Lg L f h1  u2
1 2
(23)
Where:
L f h1 L f h1 L f h1 L f h1 L f h1
L2f h1  f1  f2  f3  f4  f5 (24)
x1 x2 x3 x4 x5

1 L f h1
Lg1 L f h1   (25)
 igv x3

1 L f h1
Lg2 L f h1   (26)
 at x4
Thus, after some passages:
 f f   f
L2f h1  p  x2 1  f2  x   x1 2  f1  x   x2 1  f1  x  
 x1 x1  x2
(27)
f   f f   f f  
 x1 2  f2  x    x2 1  x1 2  f3  x    x2 1  x1 2  f4  x  
x1   x3 x3  x4 x4 
 p  f1 f 
Lg1 L f h1   x2  x1 2  (28)
 igv  x3 x3 
 p  f1 f 
Lg1 L f h1   x2  x1 2  (29)
 at  x4 x4 
It can be numerically verified that (28) and (29) are nonzero in the system state domain corresponding to normal
operation of the machine. This implies that the relative degree is well defined and in particular 𝑟 = 2.
Focusing on the temperature channel and repeating the same procedure as above one can obtain:
y2  L f h2  Lg1 h2  u1  Lg 2 h2  u2 (30)
In which:
h2 h h h h
L f h2  f1  2 f 2  2 f 3  2 f 4  2 f5 (31)
x1 x2 x3 x4 x5
Lg1 h2  0 (32)
Lg 2 h2  0 (33)
Therefore:
y2  L2f h2  Lg L f h2  u1  Lg L f h2  u2
1 2
(34)
Where:
L f h2 L f h2 L f h2 L f h2 L f h2
L2f h2  f1  f2  f3  f4  f5 (35)
x1 x2 x3 x4 x5
Lg1 L f h2  0 (36)
Lg2 L f h2  0 (37)
And again:
y2  L3f h2  Lg L2f h2  u1  Lg L2f h2  u2
1 2
(38)
Where:
5

L2f h2 L2f h2 L2f h2 L2f h2 L2f h2


L h2 
3
f1  f2  f3  f4  f5 (39)
f
x1 x2 x3 x4 x5
1 L f h2
2

Lg1 L2f h2   (40)


 igv x3
1 L f h2
2

Lg2 L h2  
2
(41)
f
 at x4
Again, it can be numerically verified that (40) and (41) are nonzero in the system state domain corresponding to
normal operation of the machine. Therefore, the relative degree of the EGT is well defined and in particular 𝑟 = 3.
Hence, the global relative degree is 𝑟 = 𝑟 + 𝑟 = 5 equal to the global order of the original system, implying the
absence of the internal dynamics. Finally, the transformed system can be written as follows:
y1   f1 CF  x    b11  x  b12  x    u1 
 
     (42)
 y2   f 2 CF  x   b21  x  b22  x    u2 
Where the suffix “CF” is for “Companion Form” and the terms 𝑏 are the elements of the B-matrix. The symbols in
(42) are defined as:
f1 CF  x   L2f h1  x  (43)
f 2 CF  x   L3f h2  x  (44)
b11  x   Lg1 L f h1  x  (45)
b12  x   Lg2 L f h1  x  (46)
b21  x   Lg1 L2f h2  x  (47)
b22  x   Lg2 L h2  x  2
f
(48)

C. Model approximation assesment


In order to overcome the insurgence of the unstable internal dynamic, the approximated formula (17) for the power
generated was introduced. In this expression the 𝛼 parameter plays a fundamental role. In order to obtain the most
precise approximation of (15), its optimal value 𝛼 was calculated through a best fitting operation on a collection of
equilibrium points. Despite this study, the approximated formula with 𝛼 = 𝛼 still introduces some error in the power
channel. The functions 𝑓 , 𝑏 and 𝑏 can be seen as affected by parametric uncertainties, where their estimated
value is given by:
f̂1 CN  f1 CN (49)
 P ˆ P

b̂11  b11  ˆ P


(50)
P

b̂12  b12  ˆ (51)


P

Hence:
ˆ T
fCF   ˆf1 CF , ˆf 2 CF  (52)

 bˆ bˆ 
[ B̂ ] =  11 12  (53)
bˆ 21 bˆ22 
Where:
f̂ 2 CF  f 2 CF (54)
b̂21  b21 (55)
b̂22  b22 (56)
This error uncertain effect on the system can be coped with by the SM controller, as long it is known to be within
certain boundaries. Identifying the upper and lower limit for the 𝛼 is then of primary importance. This can be achieved
by calculating the maximum and minimum value (respectively 𝛼 and 𝛼 ) in the system state domain of the
following function:
6

h1  x 
P  x   (57)
x1 x2
More precisely (57) is the law which allows the approximated formula to equal the correct one.

III. CONTROLLER DESIGN


According to [16] the sliding variables can be defined as:
n 1
 d 
si      yi (58)
 dt 
Where 𝑛 is the order of the ith controlled channel and 𝑦 represent the ith tracking error. As regards the GT model,
the relative degrees of the two are respectively 2 and 3, therefore one has:
s1  y1  1 y1 (59)
s2  
y 2  22 y 2  22 y 2 (60)

Where:
y1  y1  yd 1 (61)
y 2  y2  yd 2 (62)
And the terms 𝑦 are the desired power and EGT values. According to the theory, the control laws for a 2x2 MIMO
system are of the form:
u1 SM  [ Bˆ ]11
1

yd 1  1 L f h1  P ˆ P
 yd 1  
 s  
 ˆf 1 CF  k1 sat  1    [ Bˆ ]12
 1  
1

yd 2  22 L2f h2   P ˆ P
 (63)


y2 d   22 L f h2
 P ˆ P  s
 y2d  ˆf 2 CF  k 2 sat  2
 2
 

 

u2 SM  [ Bˆ ]21
1

yd 1  1 L f h1   P ˆ P
 yd 1  
 s  
 1  

 ˆf 1 CF  k1 sat  1    [ Bˆ ]221 yd 2  22 L2f h2 
 P ˆ P
 (64)


y2 d   22 L f h2
 P ˆ P 
 s  
 y2d  ˆf 2 CF  k 2 sat  2  
 2  
Where 𝜙 and 𝜙 are the boundary layer widths. Moreover, in order to satisfy the sliding conditions on each sliding
variable:
si  i for si  i
 (65)

si  i for si  i
The uncertainty compensating terms have to be of the form:
1 
k1  [ I  D]11 yd 1  1 L f h1
1  F1  d11 
  P ˆ P 
 y d 1  

 fˆ1  [ I  D]12
1
2  d12 yd 2  2  L h
2
2
f 2 y 
  2d (66)
 P ˆ P


22 L f h2
 P ˆ P
 y   fˆ 
2d


2

k2  2 (67)
In particular, as regards the additive uncertainties one has:
7

 f f 
F1  x   ˆ P   P mi n   x2 1  f 2  x   x1 2  f1  x  
 x1 x1 
 f f 
  x2 1  f1  x   x1 2  f 2  x   (68)
 x2 x1 
 f1 f   f f 
 x2  x1 2  f 3  x    x2 1  x1 2  f 4  x 
 x3 x3   x4 x4 
F2  0 (69)
While, as far as the multiplicative uncertainties are concerned, they can be effectively compensated by choosing an
appropriate D-matrix. This matrix can be defined here as:
d d12 
[ D ] =  11 (70)
 0 0 
In which:
E11 bˆ 22  E12 bˆ 21
d11  (71)
bˆ11bˆ 22  bˆ12 bˆ 21

E11 bˆ12  E12 bˆ11


d12  (72)
bˆ11bˆ 22  bˆ12 bˆ 21
Where the terms 𝐸 and 𝐸 are defined as follows:
E11  b11  bˆ11 (73)

E12  b12  bˆ12 (74)


Therefore, one has:
1 f1 f
E11  x   ˆ P   P min  x2  x1 2 (75)
 igv x3 x3
1 f1 f
E12  x   ˆ P   P mi n  x2  x1 2 (76)
 at x4 x4

IV. GOALS AND TUNING OF THE CONTROLLER


A. Goals
This paragraph is dedicated to appropriately tune the control system in order to obtain the desired performance, in
particular the main goals slightly differ from load increase to load rejection operations. In both situations, the aim is to
follow the desired power profile as quick as possible and to keep the EGT constant, however, the emergence of CCT
peaks must be avoided in the former case. In the latter, the CCT is not a critical variable instead, due to its decreasing
profile. In addition, one further aim is to exploit the same controller (designed on a 2 input – 2 output system) in
operating conditions when the IGV reach their maximum and minimum opening position (i.e. at the loss of the control
over one input).
B. Control system tuning
The control action can be conceptually divided in two phases: firstly, taking the sliding variable 𝑠 to the sliding
surface 𝑠 (𝑥̅ ) = 0 (or to its proximity) and secondly, keeping it on that surface. It is easy to notice from (58) that the
tracking error is a filtered version of the sliding variable, hence:
1. For 𝑠̇ ≠ 0 the error dynamics is roughly given by the sliding variable dynamics.
2. For 𝑠̇ = 0 the error dynamics is a linear dynamic with a time constant equal to (𝑛 − 1)/𝜆 , converging to
( )
𝑠 /𝜆 .
In particular, the parameters that can be regulated in order to tune the controller, are basically two for each channel:
𝜂 and 𝜆 . It is easy to notice from (65) that the former sets a minimum value for |𝑠̇ |, thus, a minimum value for the
“speed” of the ith-sliding variable reaching its boundary layer. The latter regulates the ith-error dynamics with respect
8

to 𝑠 , therefore it plays a crucial role when the sliding variable becomes zero or a near value. The following block
scheme clarifies the dynamical link between 𝑠 and the tracking error 𝑦 :
si 1 / ini 1 y i

1  s / i 
ni 1

Fig. 1. Structure of the closed loop error dynamics

On balance, one can say that the 𝜂 parameter substantially regulates the behavior of the system outside the
boundary layer, while 𝜆 essentially affects the behavior of the system inside the boundary layer. According to the
considerations mentioned above the numerical values for 𝜂 and 𝜂 are reported in table 1. As regards the choice of 𝜆
and 𝜆 , these parameters were not set constant, but time varying instead. Indeed, in order to obtain the desired
performance from the controller, a continuous regulation depending on the current work point, is needed. In
particular, during load increase operations the regulation of 𝜆 was designed as follows:
1    y1  10 (77)
Where the 𝜌 coefficient depends on the power tracking error according to the following law:

Fig. 2. Regulation of 𝜆 in load increase operations

While:
2  20 (78)
During load rejection operations the regulation of 𝜆 is absent, therefore one has:
1  10 (79)
While 𝜆 is regulated thanks to a multiplicative coefficient as illustrated in the following relation:
2    y 2  20 (80)
Where the coefficient 𝜓 varies as shown below:

Fig. 3. Regulation of 𝜆 in load rejection operations


9

V. SIMULATIONS
A. Description of the Gas Turbine test model
In order to test the robustness of the proposed controller, it was applied on a more sophisticated model of the gas
turbine which is used by ANSALDO ENERGIA s.p.a. for the AE94.3A machine [21]. The main differences to the model
exploited for the controller construction are the following:
 The approximated formula for the power generated (17) is replaced by the correct one (15).
 The first order linear dynamics describing the actuator behavior is replaced by more accurate models.
More detailed information and are given in [6], together with a dynamical comparison between the two models.
B. Gas turbine traditional control
The control systems nowadays implemented on gas turbine gensets are basically of the PID type as previously
mentioned. The conventional regulation basically aims at tracking the desired power generation and keeping the EGT
constant. The latter goal on the one hand is due to the bottomer heat power plant requirements [22], on the other hand,
keeping the EGT constant as much as possible during load increase operations drastically reduces CCT overshoots.
Typically, these control systems can be split in three parts: one controller dealing with the machine speed regulation
(used only in startup phases), one addressing the power regulation and a third one dedicated to the EGT control.
Ignoring the speed regulator, a basic but straightforward scheme of the control used by Ansaldo Energia S.P.A. can be
shown:
y1
y 1d
  
1 + sTLC
s 
Load Controller

u2
K EF
min

y 2d  K TL + sTTL
 s
Load Limiter

  u1
K TC + sTTC
  s
Temperature
controller
offset

y2

Fig. 4. Basic scheme of conventional PID control for gas turbines

From Fig. 4 one can notice how the desired fuel flow is obtained from a minimum selection between the power
regulator output and the temperature limiter signal. The latter is fundamental in preserving the integrity of the turbine
according to the EGT measurement. In particular, its function is to limit the fuel flow during rapid load increase in
order to keep the EGT constant as much as possible (i.e. reducing CCT peaks). Therefore, it provides a protection
function to the machine. More details on this conventional control system can be found in [6].
C. Simulation description
Two sets of simulation are presented here, namely, the system response to a step load increase (from 150 MW to 250
MW keeping the EGT constant) and to a step load rejection (from 150 MW to 50 MW). The former simulation purpose
is to show the good time response of the power channel thanks to the SM control, in spite of avoiding CCT peaks. The
latter aims at highlighting the robustness of the proposed controller even to the loss of regulation over one input (i.e.
the air flow). The numerical values of the controller parameters are listed in table I.
TABLE I
10

CONTROLLER PARAMETER NUMERICAL VALUES


Symbo
Quantity Value
l
𝜙 Power channel boundary layer 10 [MW/s2]
𝜙 EGT channel boundary layer 20 [K/s3]
Sliding variable 1 condition
𝜂 0.01 [MW/s2]
coefficient
Sliding variable 2 condition
𝜂 800 [K/s3]
coefficient
𝜆 Sliding variable 1 coefficient 2.5 [s-1]
𝜆 Sliding variable 2 coefficient 3.5 [s-1]
Estimated coefficient of the
𝛼 0.098 [W/(Pa·K)]
approximated power formula
Minimum value of coefficient of the
𝛼 0.091 [W/(Pa·K)]
approximated power formula
𝛼 Minimum value of coefficient of the 0.1 [W/(Pa·K)]
approximated power formula
The numerical values of the gas turbine model can be found in [6].
D. Step load increase
The first set of simulation is shown below:

Fig. 5. Performance comparison in load increase – Generated Power

The power dip in red indicates the load limiter intervention in Fig. 5, and his action continues until the power
reaches its desired value. By protecting the machine from overtemperature, it originates a slower dynamics for the
power channel, hence, it deteriorates its dynamical performance.

Fig. 6. Performance comparison - real and measured EGT

In order to show the effect of the “filtering action” caused by the temperature sensor, not only the measured EGTs
are shown, but also the real ones in the dashed line.
11

Fig. 7. Performance comparison in load increase – CCT

Fig. 8. Performance comparison - Combustion Chamber Pressure

Fig. 9. Performance comparison in load increase – actual air and fuel flow
12

Fig. 10. Sliding variables

From Fig. 5 it easy to appreciate the performance improvement of the power channel thanks to the SM controller,
besides, not only the power reaches the desired value significantly faster, but the CCT profile does not show any spike
(see Fig. 7). In particular, this is obtained only thanks to the Sliding Mode controller, namely, without any additional
load limiter protection. Lastly, the SM controller keeps the power and EGT steady state errors respectively less than 1
MW and 1 °C.
E. Step load rejection

Fig. 11. Performance comparison - Generated Power

Fig. 12. Performance comparison - real and measured EGT


13

Fig. 13. Performance comparison – CCT

Fig. 14. Performance comparison - Combustion Chamber Pressure

Fig. 15. Performance comparison - actual air and fuel flow

Keeping the control on the generated power in spite of the loss of regulation over the air flow, proves the robustness
of the SM controller. The performance of the two controllers are comparable in load rejection operations, nevertheless
avoiding negative power overshoot in the SM case. It can be verified again that the power steady state error is less
than 1 MW.
14

Fig. 16. Sliding variables

VI. CONCLUSIONS
A novel model based control for gas turbines aiming at enhancing their dynamical performance is proposed in this
paper. In particular, in order to cope with the complexities of gas turbine models, the attention was focused on
developing a robust nonlinear Sliding Mode controller. This controller was designed a Vth order nonlinear model
transformed via I/O FBL based on an approximated formula of the generated power, in order to avoid the emergence
of unstable internal dynamics. This model approximation was appropriately handled by the robustness of the control
system once the majorants of the errors originated by it had been defined. In addition, in order to obtain the desired
performance, a convenient tuning of the main parameters of the controller had to be carried out, depending on both
the tracking errors. Finally, the conventional PID control and the SM technique were compared simulating firstly a
step load increase and secondly a step load rejection. The power channel exhibited significantly improved performance
in the former case thanks to the model based controller, which was also able to avoid the emergence of CCT peaks
without the assistance of a safety temperature limiter. The two control systems showed similar performance in load
rejection operations, besides the SM control proved its robustness in that the control on the generated power was kept
despite the reaching of the IGV minimum opening position.

REFERENCES
[1] L. Mudan and W. Yinsong, "Power fuzzy adaptive control of gas turbine units for WGCPGS," in 2017 32nd Youth Academic Annual
Conference of Chinese Association of Automation (YAC), 2017, pp. 507-510.
[2] A. Botterud. (2017). Electricity Markets and Renewable Energy: United States vs. Europe. Available:
https://www.ethz.ch/content/dam/ethz/special-interest/itet/institute-eeh/power-systems-dam/documents/Seminar/Botterud-
ETH-20170127.pdf
[3] J. Shin, Y. Jeon, D. Maeng, J. Kim, and S. Ro, "Analysis of the dynamic characteristics of a combined-cycle power plant," Energy, vol. 27,
pp. 1085-1098, 2002.
[4] F. Mansour, A. Abdul Aziz, S. Abdel-Ghany, and H. El-Shaer, "Combined cycle dynamics," Proceedings of the Institution of Mechanical
Engineers, Part A: Journal of Power and Energy, vol. 217, pp. 247-258, 2003.
[5] M. Nagahara, Y. Yamamoto, S. Miyazaki, T. Kudoh, and N. Hayashi, "H∞ control of microgrids involving gas turbine engines and
batteries," in Decision and Control (CDC), 2012 IEEE 51st Annual Conference on, 2012, pp. 4241-4246.
[6] A. Bonfiglio, M. Invernizzi, D. Lanzarotto, A. Palmieri, and R. Procopio, "Definition of a sliding mode controller accounting for a reduced
order model of gas turbine set," in 2017 52nd International Universities Power Engineering Conference (UPEC), 2017, pp. 1-6.
[7] W. I. Rowen, "Simplified mathematical representations of heavy-duty gas turbines," Journal of engineering for power, vol. 105, pp. 865-869,
1983.
[8] F. Jurado, M. Ortega, A. Cano, and J. Carpio, "Neuro-fuzzy controller for gas turbine in biomass-based electric power plant," Electric Power
Systems Research, vol. 60, pp. 123-135, 2002/01/28/ 2002.
[9] S. Camporeale, L. Dambrosio, and B. Fortunato, "One Step Ahead Adaptive Control for Gas Turbine Power Plants," in ASME 1999
International Gas Turbine and Aeroengine Congress and Exhibition, 1999, pp. V004T04A004-V004T04A004.
[10] E. Najimi and M. H. Ramezani, "Robust control of speed and temperature in a power plant gas turbine," ISA transactions, vol. 51, pp. 304-
308, 2012.
[11] M. RAHNAMA, A. GHAFFARI, and H. GHORBANI, "Constrained Model Predictive Control Implementation for a Heavy-Duty Gas
Turbine Power Plant."
[12] K. Jordal, M. Assadi, and M. Genrup, "Variations in Gas-Turbine Blade Life and Cost due to Compressor Fouling–A Thermoeconomic
Approach," International Journal of Thermodynamics, vol. 5, pp. 37-47, 2002.
[13] D. Rouwenhorst and J. Hermann, "Precursor for thermoacoustic stability in annular combustion systems, based on output-only system
identification," in EVI-GTI and PIWG Joint Conference on Gas Turbine Instrumentation, 2016, pp. 1-5.
View publication stats

15

[14] J. Lepers, W. Krebs, B. Prade, P. Flohr, G. Pollarolo, and A. Ferrante, "Investigation of thermoacoustic stability limits of an annular gas
turbine combustor test-rig with and without Helmholtz resonators," ASME Paper No. GT2005-68246, 2005.
[15] F. Saccomanno, Electric power systems: analysis and control: Wiley Online Library, 2003.
[16] J.-J. E. Slotine and W. Li, Applied nonlinear control vol. 199: Prentice hall Englewood Cliffs, NJ, 1991.
[17] G. Park and Z. Gajic, "A Simple Sliding Mode Controller of a Fifth-Order Nonlinear PEM Fuel Cell Model," IEEE Transactions on Energy
Conversion, vol. 29, pp. 65-71, 2014.
[18] C. Evangelista, F. Valenciaga, and P. Puleston, "Active and Reactive Power Control for Wind Turbine Based on a MIMO 2-Sliding Mode
Algorithm With Variable Gains," IEEE Transactions on Energy Conversion, vol. 28, pp. 682-689, 2013.
[19] R. Shahnazi, H. M. Shanechi, and N. Pariz, "Position Control of Induction and DC Servomotors: A Novel Adaptive Fuzzy PI Sliding Mode
Control," IEEE Transactions on Energy Conversion, vol. 23, pp. 138-147, 2008.
[20] W. Wen-Jieh and C. Jenn-Yih, "Passivity-based sliding mode position control for induction motor drives," IEEE Transactions on Energy
Conversion, vol. 20, pp. 316-321, 2005.
[21] A. E. S.P.A. AE94.3A: The proven value. Available: https://www.ansaldoenergia.com/business-lines/new-units/gas-turbines/ae94-3a
[22] H. I. H. Saravanamuttoo, G. F. C. Rogers, and H. Cohen, Gas turbine theory: Pearson Education, 2001.

Potrebbero piacerti anche