Sie sind auf Seite 1von 8

IDENTIFICATION OF POWER SYSTEM MODEL PARAMETERS FOR SLIDING MODE BASED LOAD-

FREQUENCY CONTROL
Krešimir Vrdoljak, Student Member, IEEE, Tamara Radošević, Student Member, IEEE, Nedjeljko Perić,
Senior Member, IEEE

optimal control [12,13], adaptive control [14,15],


Abstract: The most studies of load-frequency control in
variable structure control [16,17] and other.
power systems are done using linearized power system
models. Those models may have nonlinearities within Nevertheless, LFC algorithms used in real power
turbines and their governors, but generator dynamics systems are still mostly PI type [18,19], although LFC
and nonlinearities in tie-lines are mostly neglected. based on PI control algorithm has long settling time
Generally, those models are only valid in the vicinity of and relatively large overshoot [20].
the system’s operating point. In this paper, the behavior What is the reason for the prevalent usage of PI
of an advanced load-frequency controller is tested on algorithms in real power systems in relation to
nonlinear power system model. The following procedure advanced control techniques? Could it be that
is used: firstly, by analyzing structure and response simplified and linearized models used in the studies of
signals of the nonlinear model, parameters of the
advanced methods are not adequate, and good results
substitute linear model are obtained. Then, from the
substitute linear model, sliding mode based load- obtained on simulations do not replicate in real power
frequency controller is synthesized. Finally, the systems? In this paper we’ll try to find out if that’s true
controller’s behavior is validated on the nonlinear model. for one advanced LFC algorithm. The algorithm is
taken from [21], where it is designed for linearized
Index Terms: load-frequency control, sliding mode power system model.
control, identification, optimization, SimPowerSystems.
The used LFC control algorithm is derived from
discrete-time sliding mode control theory. Sliding
mode control (SMC) is a robust control technique that
I. INTRODUCTION
shows very good behavior in controlling systems with
external disturbances and parameter variations [22].
Power systems are complex, nonlinear systems,
Power systems are such systems. In SMC, system
which are composed of interconnected subsystems or
closed-loop behavior is determined by a sub-manifold
control areas (CAs). Every CA consists of coherent
in the state space. While the state is on the manifold,
group of generators and it is connected with other CAs
system invariance to particular uncertainties and
by tie-lines. In each CA, load-frequency control (LFC)
parameter variations is guaranteed.
must keep the frequency at its nominal value and must
A nonlinear model is made using Matlab’s
maintain area's scheduled active power flow
SimPowerSystems Toolbox [23].
interchange with the neighbor CAs [1]. Frequency and
The brief outline of the paper is as follows: Section
power flow deviate from their nominal or contracted
II presents linear and nonlinear power system models,
values due to differences in generated and consumed
while in Section III system identification technique is
active power in the system. Power system is usually
used to find a correlation between parameters of the
modeled as interconnection of a few CAs, where each
two models. Section IV describes sliding mode based
CA consists of either one substitute thermal or hydro
LFC algorithm and shows results obtained by using
power plant [2-5]. In each area there is an appropriate
that algorithm and linear and nonlinear power sstem
LFC controller.
models.
LFC algorithms are tested on power system
models, because it is inappropriate to test them on real
II. POWER SYSTEM MODELS
power system. It is well known that linearized models
are valid only in the vicinity of the operating point.
Power system models used in LFC algorithms
Therefore, they trustworthily represent real power
studies usually consist of a few interconnected CAs.
system dynamics only for close-to-normal system
An example model consisting of four CAs in ring-type
operation. Additionally, linear models have constant
interconnection will be used here (Fig. 1). Each CA in
parameters, while parameters of real power systems
the model is represented with one substitute power
constantly vary in time due to their dependency on the
plant and has one associated load-frequency controller.
operating condition. That condition is determined by
It is supposed that substitute power plants in CA1 and
size and characteristic of consumption and by number
CA3 are thermal power plants, while those in CA2 and
and characteristics of power plants engaged in LFC.
CA4 are hydro power plants.
Throughout the recent decades many different LFC
algorithms were studied and they showed very good
A. A Linearized Model (LIN-model)
results on linearized power system models [6]. Those
algorithms are based on various methods, such as
For most studies of LFC algorithms, linear power
robust control theory [7,8], fuzzy logic [2,9], neural
system models are used because only small changes of
networks [4,5], model predictive control [10,11],

540 978-1-4244-3861-7/09/$25.00 ©2009 IEEE


controller and di(t) presents differences between
Control Area 1 Control Area 2 produced and consumed active power.
(thermal) (hydro) For a CA represented with thermal power plant
(Fig. 2), system matrices and state vector in (1) are
given in (3), while for a CA represented with hydro
power plant (Fig. 3), they are given in (4). Description
of parameters and variables from (3) and (4) are given
Control Area 3 Control Area 4 in Tables 1 and 2.
(thermal) (hydro) All the elements of matrices Aij in (1) are equal to
zero, except the element at position (1,2) which is
Fig. 1 – Interconnection of 4 control areas equal to −KS. Dimensions of those matrices depend
whether they are used to describe a connection with
load are expected during normal system operation. A thermal or hydro CA.
linear model usually consists of just one substitute An area control error (ACE) signal is introduced as
power plant per CA. However, studies concerning a quantitive measure of CA’s deviation from the
deregulation effects on LFC may have several proposed behavior. The goal of LFC is to compensate
substitute power plants modeled within one CA [3,12]. for area’s ACE deviations. ACE is defined as:
Linear dynamics of a CA can be described with the
following equation: ACE (t ) = K B Δf (t ) + ΔPtie (t ) , (2)

N where frequency bias parameter, KB, is tuned in a way


xi (t ) = Aii xi (t ) + ∑ Aij x j (t ) + Bi ui (t ) + Fi d i (t ) , (1) that ensures ACE different than zero only when
j =1, j ≠ i
disturbances occur within the area. Thus, if a
disturbance happens elsewhere in the interconnection,
where xi(t) is system state, ui(t) input and di(t) area’s ACE will be zero and area’s LFC will not
disturbance vector. ui(t) is computed within LFC compensate for that disturbance.

1
R
ΔPd

- 1 Δx g 1 ΔPg - KP 2π
Δδ i

+
1 + sTG 1 + sTT + - 1 + sTP Δf s

u ΔPtie Ks
LFC ∑ 2π (Δδ
j
i − Δδ j )

Fig. 2 – Linearized model of a control area with substitute thermal power plant

1
R
ΔPd

- 1 Δxgh 1 + sTR Δxg 1 − sTW ΔPg - KP 2π


Δδ i

+
1 + sT1 1 + sT2 1 + 0.5sTW + - 1 + sTP Δf s

u ΔPtie Ks
LFC ∑ 2π (Δδ
j
i − Δδ j )

Fig. 3 – Linearized model of a control area with substitute hydro power plant

541
⎡ 1 KP KP ⎤ Table 2. List of parameters of the linear model
⎢ −T −
TP TP
0 ⎥
Symbol Parameter Unit
⎢ P ⎥ KP power system gain Hz/p.u.MW
⎢ N ⎥
⎢ ∑ KS 0 0 0 ⎥ TP power system time constant s
TW water starting time s
A=⎢ ⎥,
j

⎢ 1 1 ⎥ T1, T2 hydro governor time constants s


⎢ 0 0 − ⎥ hydro governor transient droop time
⎢ TT TP ⎥ TR
constant
s
⎢ 1 1⎥ (3) TG thermal governor time constant s
⎢− 0 0 − ⎥ TT turbine time constant s
⎣⎢ RTG TG ⎦⎥
KS interconnection gain between CAs p.u.MW
⎡0⎤ ⎡ K Pi ⎤ R speed droop due to governor action Hz/p.u.MW
⎢0⎥ ⎢− T ⎥ ⎡ Δf ( t ) ⎤
KB frequency bias factor p.u.MW/Hz
⎢ ⎥ ⎢ Pi ⎥ ⎢ ΔPtie (t ) ⎥⎥
B = ⎢ 0 ⎥ , F = ⎢ 0 ⎥ , x(t ) = ⎢ . power systems.
⎢ ⎥ ⎢ ⎥ ⎢ ΔPg (t ) ⎥ The model of a CA consists of one substitute power
⎢1⎥ ⎢ 0 ⎥ ⎢ ⎥
⎡ −1⎢ T ⎥ − K P ⎢ K P0 ⎥ ⎣⎢ Δxg (t ) ⎦⎥⎤ plant, transformer, transmission lines, power system
⎢ T⎣ G ⎦ T ⎣ T ⎦ 0 0 ⎥ stabilizer, loads and LFC controller. Substitute power
⎢ P P P ⎥ plant consists of a synchronous machine, thermal or
⎢N ⎥ hydro turbine, speed governor and excitation system.
⎢∑ K S 0 0 0 0 ⎥
Different types of loads can be switched on or off
⎢ j ⎥
⎢ 2TR −2 T2 + TW TR − T1 ⎥ using breakers, thus representing disturbances in the
A=⎢ 0 ⎥, system.
⎢ T1T2 R TW 0.5T2TW 0.5T1T2 ⎥ The model of a CA built in SimPowerSystems is
⎢ −T −1 T1 − TR ⎥ shown in Fig. 4. Parameters of all 4 CAs in the
⎢ R 0 0 ⎥
⎢ T1T2 R T2 T1T2 ⎥ interconnection are given in Table 3.
⎢ −1 SMC based LFC algorithm that will be tested here
−1 ⎥
⎢ 0 0 0 ⎥ was developed for linear power system model [21].
⎣⎢ T1 R T1 ⎦⎥ Therefore, the controller will not be designed directly
⎡ 0 ⎤ for SPS-model, but to its substitute LIN-model.
⎢ ⎥ Parameters of the substitute model can be identified by
⎢ 0 ⎥ ⎡ −KP ⎤
⎢ T ⎥ ⎡ Δf (t ) ⎤ analyzing structure and step responses of SPS-model’s
⎢ −2TR ⎥ ⎢ ΔP (t ) ⎥ subsystems.
⎢ ⎥ ⎢ P ⎥

T1T2 ⎥ ⎢ 0 ⎥ ⎢ tie ⎥

B=
⎢ TR ⎥
, F=⎢ ⎥ , x (t ) = ⎢ ΔPg (t ) ⎥ . Table 3. Values of the parameters of the nonlinear model
⎢ 0 ⎥ ⎢ ⎥
⎢ ⎥
⎢ ⎥ ⎢ Δxg (t ) ⎥ Parameter CA1 CA2 CA3 CA4
T T
⎢ 1 2 ⎥ 0 ⎢ Δx (t ) ⎥
⎢ 1 ⎥ ⎢ ⎥ ⎣ gh ⎦ nominal frequency fn [Hz] 50
⎢⎣ 0 ⎥⎦
⎢ ⎥ nominal power Sn [MVA] 900 900 1200 1100
⎣⎢ T1 ⎦⎥ (4) nominal voltage Un [kV] 20 15 21 18
d-axis synchronous reactance Xd
B. A Nonlinear Model (SPS-model) 1.8 1.3 2 1
[p.u.]
Table 1. List of state variables of the linear model d-axis transient reactance Xd' [p.u.] 0.3 0.3 0.35 0.36
Symbol Variable Unit d-axis subtransient reactance Xd''
0.25 0.25 0.2 0.2
f(t) frequency Hz [p.u.]
Pg(t) generator output power p.u.MW q-axis synchronous reactance Xq
1.7 0.45 1.9 0.58
[p.u.]
xg(t) governor valve position p.u.
q-axis transient reactance Xq' [p.u.] 0.55 - 0.7 -
xgh(t) governor valve servomotor position p.u
q-axis subtransient reactance Xq''
Ptie(t) tie-line active power p.u.MW 0.25 0.24 0.2 0.33
[p.u.]
Pd(t) load disturbance p.u.MW
leakage reactance Xl [p.u.] 0.2 0.18 0.15 0.09
δ(t) rotor angle rad
d-axis transient open-circuit time
f(t) frequency Hz 8 5 10 5
constant Tdo' [p.u.]
ACE (t) area control error p.u.MW
d-axis subtransient open-circuit
0.03 0.02 0.02 0.05
time constant Tdo'' [p.u.]
q-axis transient open-circuit time
To validate LFC algorithms, more detailed and 0.4 - 1 -
constant Tqo' [p.u.]
complex model of power system is needed. The model q-axis subtransient open-circuit
is built in Matlab’s SimPowerSystems Toolbox, which 0.05 0.05 0.04 0.07
time constant Tqo'' [p.u.]
is a tool within Simulink, developed for modeling and stator resistance Rs [p.u.] 0.002 0.01 0.003 0.008
simulating of electrical power systems. It provides coefficient of inertia H [s] 6 2.5 9 3.5
models and libraries of many components used in pole pairs p 2 18 1 20

542
Fig. 4 – Nonlinear model of a control area in SimPowerSystem

III. IDENTIFICATION OF SUBSTITUTE LINEAR A comparison of linear and nonlinear turbine


MODEL PARAMETERS model responses, with equal values of TW, is shown in
Fig. 6. The reason for different responses is turbine
Remark 1: Frequency in LIN-model is in absolute gain in Fig. 5, which is associated with nonlinearity
value (Hz), while in SPS-model it is in relative value that limits gate’s positions.
(p.u.). Parameters T1, T2 and TR are valid only for hydro
Parameters KB can be obtained from SPS-model by power plants. They can be identified by analyzing a
applying individual step disturbances to all four CAs at nonlinear model of hydro turbine’s governor, which is
various time instants. Then, by analyzing system’s shown in Fig. 7. After neglecting nonlinearities and
frequency and tie line power responses, KB is chosen using (5), a transfer function from the reference power
such that it ensures ACE equal to zero for all (output signal of LFC controller) to the governor’s
disturbances occurring in neighbor areas. Because KB output becomes:
is not equal for all three cases of outer disturbances, its
average value is calculated. Furthermore, due to the 1 1/(gmax -gmin)
uncertainty of self regulating effect of the load, in real Xg
u
2

1
turbine Product 2
power systems KB may be chosen slightly higher, so gain
Math
Function Pg

LFC will accentuate the effect of the speed governors’ 1


s 1/Tw 1

control and not counteract it [1]. Therefore, 10%


higher values than the average value is chosen for KB. Fig. 5 – Hydro turbine’s nonlinear model
Parameter R is set as a parameter of both, thermal
and hydro governors within SPS-model. Therefore, it Area 2
needs not to be identified. Because of Remark 1 0.02
Linear
0
parameter R from SPS-model must be multiplied with Nonlinear
Δ Pg [p.u.MW]

-0.02
the nominal frequency f n = 50 Hz to obtain its value -0.04

for LIN-model. -0.06

Parameter TW is set as a parameter of hydraulic -0.08

turbine within SPS-model. Therefore, it needs not to be


-0.1
0 5 10 15 20 25 30
time [min]
identified. Nevertheless, a comparison of hydraulic Area 4
0.1
turbine subsystems of LIN- and SPS-model is given 0.08
Linear

here. Nonlinear hydro turbine model is shown in Fig. Nonlinear


Δ P g [p.u.MW]

0.06

5. In LIN-model its non-minimum phase linear 0.04

approximation is: 0.02

-0.02
1 − sTW 0 5 10 15 20 25 30

ΔPg = Δxg . (5) time [min]

1 + 0.5sTW Fig. 6 – Identification of water starting time constant

543
R p k a ( ki + k p s )
F5

Δxg
F4

s ( k a + s + s 2 ta )
= , (6) F3
1

R p ka (ki + k p s )(1 − sTW )


Pg

ΔPref
1+
1 1 1 1
1 F2
T4.s+1 T3.s+1 T2.s+1 T1.s+1

s (ka + s + s 2 ta )(1 + 0.5sTW )


Xg
Stage 4 Stage 3 Stage 2 Stage 1
2
W

Fig. 9 –Steam turbine’s nonlinear model


which can be simplified to:
Area 1

Δxg (1 + TR s ) 0.8
= . (7) 0.78
ΔPref (1 + T1 s )(1 + T2 s )

Pg [p.u.MW]
0.76

0.74
0.72

Simplification is done by cancellation of the nearby 0.7 Linear


Nonlinear
pole and zero pair, and also by approximating 0.68
2.4 2.6 2.8 3 3.2 3.4 3.6
time [min]
(1 + Ti s )(1 + T j s ) with 1 + (Ti + T j ) s , for the two Area 3
1
smallest time constants. Identification results are 0.98

shown in Fig. 8. 0.96

Pg [p.u.MW]
Parameter TT is valid only for thermal power 0.94

plants. It can be identified from the response of steam 0.92

Linear
turbine’s output power to a step change in governor’s 0.9
Nonlinear
0.88
gate opening. Differences between turbines used in 2.4 2.6 2.8 3
time [min]
3.2 3.4 3.6

CA1 and CA3 are in time constants and torque Fig. 10 – Identification of steam turbine time constant
fractions of high, intermediate and low pressure
sections. Steam turbine’s model is shown in Fig. 9 and approximated with servo motor time constant, Tsm,
its linear approximation is: which is set as a parameter of thermal governor
subsystem in SPS-model. The subsystem is shown in
1 Fig. 11, while identification results are shown in Fig.
ΔPg = Δxg . (8) 12. Both thermal power plants have identical
1 + sTG
governors.
To identify parameters KP and TP, an alternative
Steam turbine’s time constant identification results are
form of a transfer function describing power system
shown in Fig. 10. dynamics is used:
Parameter TG is valid only for thermal power
plants. With nonlinearities neglected, it can be Δf Δf 1
= = , (9)
ΔPΣ ΔPg − ΔPtie − ΔPd D + Ms
kp
servo-
1
wref
Proportional

ki 1/s
motor
ka
speed
limit position
1
where D is area’s damping coefficient and M is area’s
1
3 ta.s+1 s
Integral Xg
we
kd.s
td.s+1 Pref
1
Derivative 2 Servo -motor
Pref speed limits
Pe 1 1
Rp 2 1/R 1/Tsm s 1
4 Tsr.s+1
Wref Xg
Dead Zone Transfer Fcn Servo -motor
permanent
position
droop 3 1/R
W
Fig. 7 – Hydro turbine’s governor nonlinear model
Fig. 11 – Steam turbine’s governor nonlinear model
Area 2

0.74 Area 1
Linear 0.8
Nonlinear Linear
0.72 0.78
Nonlinear
x g [p.u.]

0.7 0.76
Xg [p.u.MW]

0.74
0.68
0.72
0.66
0.7
0.64
0 10 20 30 40 50 60 70 80 90 100 0.68
time [min] 9.9 10 10.1 10.2 10.3 10.4 10.5
time [min]
Area 4
Area 3
0.64 Linear
0.9
Nonlinear
0.62
Xg [p.u.MW]
xg [p.u.]

0.6
0.85
0.58

0.56 Linear
0.8
Nonlinear
0.54 29.9 30 30.1 30.2 30.3 30.4 30.5
0 10 20 30 40 50 60 70 80 90 100
time [min] time [min]

Fig. 8 – Identification of hydro governor’s parameters Fig. 12 – Identification of thermal governor’s parameters

544
inertia constant [7]. Damping coefficient is identified Snj K Sji = S ni K Sij , (11)
from isolated CA’s frequency response to step changes
in generation and consumption. Because of Remark 1, where are CAs’ nominal powers, Sn, given in Table 3.
it is D = Dident / f n . Since there is only one synchronous Identified parameters slightly differ from (11), the
machine per CA, area’s inertia constant can be reason for that are losses in transmission lines.
computed as M = 2 H / f n , where H is machine’s Values of all identified LIN-model parameters are
coefficient of inertia, which is a set parameter of SPS- shown in Table 4.
model. Once all its parameters are identified, LIN-model is
Parameters KP and TP can be computed from built and its responses are compared to those of SPS-
parameters M and D as: model. The results of ACE signal comparison, without
LFC, are shown in Fig. 14. It can be seen from the
1 M figure that responses of LIN- and SPS- models are
KP = , TP = . (10) similar. Therefore, it seems reasonable to use LIN-
D D model for LFC algorithm computation.
Remark 2: LIN-model parameters were mostly
Identification results are shown in Fig. 13. Two
identified on a single CAs, without interconnections
step signals were applied to SPS-model. The first one
with other CAs. Therefore, the models’ responses did
(at time t = 0 ) in the reference of active power, and not coincidence so good in the case of interconnected
the second one (at time t = 1000 s) by switching on power system. To overcome that, parameters KPi from
additional load. From Fig. 13 it can be seen that system Table 4 were multiplied by 2, while parameters R were
dynamics in SPS-model are of higher order than those divided by 2, which resulted in very good coincidence,
in LIN-model. There are significant oscillations of as shown in Fig. 14.
frequency after the change in load. Although, static
gains K P are well identified. IV. SLIDING MODE BASED LFC
Parameter KSij can be identified from the tie line
active power deviations between i-th and j-th CA, LFC algorithm is based on SMC. SMC is a control
when disturbance occurs in one of those CAs. Because technique appropriate for time-variant systems with
of different nominal powers in areas, it is: external disturbances. Power systems are those kind of

-3
x 10 Area 1
Area 1
0.03 10
Linear
ACE [p.u.MW]

Nonlinear
0.02 5
Δ f [Hz]

0.01
0

0 Linear
-5 Nonlinear

-0.01 0 20 40 60 80 100 120


0 2 4 6 8 10 12 14 16 time [min]
time [min]
-3
x 10 Area 2
Area 2
0.1 5

0.05
ACE [p.u.MW]

0
0
Δ f [Hz]

-5
-0.05

-0.1 Linear
Linear -10
Nonlinear
Nonlinear
0 20 40 60 80 100 120
0 2 4 6 8 10 12 14 16
time [min]
time [min]
-3 Area 3
x 10
Area 3
0.03 5
Linear
Nonlinear
ACE [p.u.MW]

0.02 0
Δ f [Hz]

0.01
-5

0 Linear
-10
Nonlinear
-0.01 0 20 40 60 80 100 120
0 2 4 6 8 10 12 14 16 time [min]
time [min]
-3 Area 4
x 10
Area 4
0.1
10
Linear
0.05 Nonlinear
ACE [p.u.MW]

5
0
Δ f [Hz]

-0.05 0

-0.1 Linear
Nonlinear -5
0 20 40 60 80 100 120
0 2 4 6 8 10 12 14 16
time [min]
time [min]

Fig. 14 – Comparison of LIN- and SPS- model


Fig. 13 – Identification of CA’s dynamics parameters

545
Table 4. Identified values of the parameters of LIN-model Ĝ21 = S1G11 − S1G12 S 2 −1 S1 + S2 G21 − S2 G22 S2 −1 S1 , while
Parameter CA1 CA2 CA3 CA4 x and d are estimated state and disturbance,
KP 2.643 2.645 2.617 2.551 respectively.
TP 31.717 13.223 40.098 17.855 Advances of SMC based LFC (SMCLFC) for LIN-
TW - 2.67 - 1.5 model are shown in [21]. Here, system behavior with
T1 - 197.457 - 217.518 two sets of SMCLFC parameters is tested. The first set
T2 - 0.357 - 0.423 is obtained directly from SPS-model, while the second
TR - 11.076 - 9.474 set is obtained from its substitute LIN-model. Both sets
TG 0.15 - 0.18 -
are obtained using an optimization technique from
TT 6.5 - 8 -
[21]. The results are shown in Fig. 15. It can be seen
R 2.5 2.5 2.5 2.5
KB 0.43 0.47 0.39 0.48
that the system is stable for the first set of parameters,
j while it is unstable for the second set. Therefore,
KSij advanced control algorithms for LFC, such is SMC,
1 2 3 4
1 - 2.841 2.278 0 cannot be tested only on LIN-models; they should also
2 2.873 - 0 2.230 be verified on detailed nonlinear models.
i
3 1.672 0 - 1.341
4 0 1.841 1.523 - V. CONCLUSIONS

systems. Additionally, SMC is chosen as LFC In this paper, a comparison of linear and nonlinear
algorithm due to its robustness, which should ensure power system model behavior with sliding mode based
good system behavior despite differences of LIN- and load-frequency controller is done. Firstly, a procedure
SPS- model. for obtaining substitute linear model parameters from
SMC can be based on state or output signal. Output the responses of a nonlinear model is given. Then,
based SMC can not be used for non-minimum phase optimal controller’s parameters are obtained directly
system [24]. Since hydro power plants have non- from the nonlinear model and also from its substitute
minimum phase behavior, state based SMC is used linear model. Nonlinear power system model with
here. Not all system state can be measured; therefore, controller’s parameters obtained from the substitute
unmeasurable state variables are estimated. Estimation linear model has unstable behavior. It is shown that
method used here is described in detail in [25]. good behavior of an advanced controller achieved on
SMC design procedure consists of two major steps: linear power system model doesn’t necessary replicate
1) selection of a sub manifold in state space with on nonlinear power system model.
desired system behavior and 2) computation of a Therefore, the focus of the future research should
control law that will force system's trajectory to reach be in improving SMC based algorithm to show good
and stay on that manifold. behavior with both, linear and nonlinear power system
A discrete time sliding mode control is used, models.
because in real power systems LFC is implemented to
send the control signal to the power plant every few
seconds [18]. To compute the control law, system (1) 5
x 10
-4
LIN-model

firstly needs to be converted to the discrete-time 0

equivalent and then transformed into the regular form:


ACE [p.u.MW]

-5

-10

⎡ x1 (k + 1) ⎤ ⎡ G11 G12 ⎤ ⎡ x1 (k ) ⎤ -15

⎢ x (k + 1) ⎥ = ⎢G ⎥⎢ ⎥+
⎦ ⎣ 21 G22 ⎦ ⎣ x2 (k ) ⎦
-20
⎣ 2 0 0.2 0.4 0.6 0.8 1 1.2

(12) time [min]

⎡ 0 ⎤ ⎡W ⎤ 0.15
SPS-model (parameters from LIN-model)

+ ⎢ ⎥ u (k ) + ⎢ 1 ⎥ d (k ).
⎣H2 ⎦ ⎣W2 ⎦
0.1
ACE [p.u.MW]

0.05

System trajectory’s dynamics of reaching the -0.05

manifold σ (k ) = [ S1 S2 ][ x1 (k ) x2 (k ) ] , are given


T -0.1

0 0.2 0.4 0.6 0.8 1 1.2

as: time [min]

SPS-model (parameters from SPS-model)


σ (k + 1) = λσ (k ) . (13) 0.01

0.005
ACE [p.u.MW]

0
Combining (12) and (13), the control law can be -0.005
computed as: -0.01

-0.015
0 1 2 3 4 5 6 7

u = − Hˆ 2 −1 ⎡⎣ (λ − Gˆ 22 )σ (k ) − Gˆ 21 x1 (k ) − Wˆ2 d (k ) ⎤⎦ ,(14) time [min]

Fig. 15 – SMCLFC for LIN- and SPS-model


where are Ĥ 2 = S2 H 2 , Ĝ22 = S1G12 S 2 −1 + S2 G22 S 2 −1 ,

546
Go to Table of Content

REFERENCES [17]. Yang M.S. Lu H.C. Sliding mode load-frequency


controller design for dynamic stability enhancement of
[1]. UCTE. Operational Handbook - Load-Frequency large-scale interconnected power systems. Proceedings
Control and Performance. http://www.ucte.org, 2004. of the “IEEE International Symposium on Industrial
[2]. Cam E. Application of fuzzy logic for load frequency Electronics”, Bled, Slovenia, 1999, pp. 1316-1321.
control of hydroelectrical power plants, Energy [18]. Rebours Y.G. Kirschen D.S. Trotignon M. Rossignol S.
Conversion and Management 48 (4) (2004). p. 1281- A Survey of Frequency and Voltage Control Ancillary
1288. Services-Part I: Technical Features, IEEE Transactions
[3]. Bevrani H. Mitani Y. Tsuji K. Bevrani H. Bilateral on Power Systems 22 (1) (2007). p. 350-357.
based robust load frequency control, Energy Conversion [19]. Stojković B. An original approach for load-frequency
and Management 56 (7-8) (2005). p. 1129-1146. control - the winning solution in the Second UCTE
[4]. Hemeida A.M. Wavelet neural network load frequency Synchronous Zone, Electric Power Systems Research
controller, Energy Conversion and Management 46 (9- 69 (1) (2004). p. 59-68.
10) (2005). p. 1613-1630. [20]. Al-Musabi N.A. Design of Optimal Variable Structure
[5]. Shayeghi H. Shayanfar H.A. Application of ANN Controllers: Applications to Power System Dynamics,
technique based next term on µ-synthesis to load Masters Thesis, King Fahd University of Petroleum and
frequency control of interconnected power system, Minerals, Dhahran, Saudi Arabia, 2004.
Electrical Power & Energy Systems 28 (7) (2006). p. [21]. Vrdoljak K. Perić N. Mehmedović M. Optimal
503-511. parameters for sliding mode based load-frequency
[6]. Ibraheem Kumar P. Kothari D.P. Recent Philosophies control in power systems. Proceeding of the
of Automatic Generation Control Strategies in Power “International Workshop on Variable Structure Systems
Systems, IEEE Transactions on Power Systems 20 (1) (VSS '08)”, Antalya, Turkey, 2008, pp. 331-336.
(2005). p. 246-357. [22]. Edwards C. Spurgeon S.K. Sliding Mode Control:
[7]. Bevrani H. Hiyama T. Robust decentralized PI based Theory and Applications. Taylor and Francis, 1998.
LFC design for time delay power systems, Energy [23]. The MathWorks Inc. SimPowerSystems User’s Guide,
Conversion and Management 49 (2) (2008). p. 193-204. 2003.
[8]. Ngamroo I. Taeratanachai C. Dechanupaprittha S. [24]. Monsees G. Discrete-Time Sliding Mode Control, PhD
Mitani Y. Enhancement of load frequency stabilization thesis, Delft University of Technology, Netherlands,
effect of superconducting magnetic energy storage by 2002.
static synchronous series compensator based on H∞ [25]. Vrdoljak K. Perić N. Petrović I. Estimation and
control, Energy Conversion and Management 48 (4) Prediction in Load frequency Control “International
(2006). p. 1302-1312. Conference on Power Engineering, Energy and
[9]. Lee H.J. Park J.B. Joo Y.H. Robust load-frequency Electrical Drives (POWERENG 2007)”, Setubal,
control for uncertain nonlinear power systems: A fuzzy Portugal, 2007, pp. 785-790
logic approach, Information Sciences 176 (23) (2006).
p. 3520-3537.
[10]. Atić N. Feliachi A. Rerkpreedapong D. CPS1 and CPS2 Krešimir Vrdoljak received his B. Sc. degree in Electrical
Compliant Wedge-Shaped Model Predictive Load Engineering from the University of Zagreb, Croatia in 2003.
Frequency Control. IEEE Power Engineering Society He has been a research assistant with the Department of
General Meeting, 2004, pp. 855-860. Control and Computer Engineering at Faculty of Electrical
[11]. Venkat A.N. Hiskens I.A. Rawlings J.B. Wright S.J. Engineering and Computing in Zagreb since then. His
Distributed Output Feedback MPC for Power System research interests are in advanced control techniques applied
Control. Proceedings of the “45th IEEE Conference on to power systems. He is a student member of IEEE Power &
Decision and Control”, San Diego, USA, 2006. Energy and Control Systems societies.
[12]. Demiroren A. Zeynelgil H.L. GA application to
optimization of AGC in three-area power system after Tamara Radošević received her B. Sc. degree in Electrical
deregulation, Electrical Power & Energy Systems 29 (3) Engineering from the University of Zagreb, Croatia in 2008.
(2007). p. 230-240. She has been a research assistant with the Department of
[13]. Parisses C. Asimopoulos N. Fessas P. Decentralized Control and Computer Engineering at Faculty of Electrical
Load-Frequency Control of a Two-Area Power System Engineering and Computing in Zagreb since then. Her
via Linear Programming and Optimization Techniques. research interests are analysis, modeling and control of
Proceedings of “5th International Conference on power systems.
Technology and Automation”, Thessaloniki, Greece,
2005, pp. 204-209. Nedjeljko Perić received his B.Sc., M.Sc. and Ph.D. degrees
[14]. Chang-Chien L.R. Cheng J.S. The Online Estimate of in Electrical Engineering from the Faculty of Electrical
System Parameters for Adaptive Tuning on Automatic Engineering and Computing, University of Zagreb, Croatia,
Generation Control. “International Conference on in 1973, 1980, and 1989, respectively. From 1973 to 1993 he
Intelligent Systems Applications to Power Systems”, worked at the Institute of Electrical Engineering of the
Niigata, Japan, 2007, pp. 1-6. Končar Corporation, Croatia. In 1993 he joined the
[15]. Zribi M. Al-Rashed M. Alrifai M. Adaptive Department of Control and Computer Engineering at FER
decentralized load frequency control of multi-area Zagreb, firstly as an associate professor and since 1997 as a
power systems, Electrical Power & Energy Systems 27 full professor. His current research interests are in the fields
(8) (2005). p. 575-583. of process identification and advanced control techniques.
[16]. Alrifai M. T. Decentralized controllers for power Prof. Perić serves as the Chairman of KoREMA, the Croatian
system load frequency control, ICGST International Society for Communications, Computing, Electronics,
Journal on Automatic Control and System Engineering Measurements and Control. Prof. Perić is a Fellow of
5 (2) (2005). p. 1-16. Croatian Academy of Engineering.

547

Das könnte Ihnen auch gefallen