Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Optimal Design of Power System Stabilizer for Multimachine Power System Using Differential Evolution
Algorithm
Ashik Ahmed1, B. M. Ruhul Amin2, Khandaker Ziaul Islam2, and Shah Md. Aminul Islam3
Dept. of EEE, IUT, Bangladesh, 2Dept. EEE, BUBT, Bangladesh, 3IMCT Division, UGC of Bangladesh
Email: ashikhmd@yahoo.com
n this modern era most of the power plants are located far
from the consumer units. Large numbers of generating units
are interconnected in order to supply ceaseless power to the
consumers at the far end. So the power industry is trying to
implement newer approaches in order to maximize power
transfer among different areas in a stable manner. The
transmission networks are usually overdesigned to keep the
operation of the overall system within safe limit considering
the constraints of thermal limit, rotor angle stability limit and
voltage stability limit [1].
The simultaneous tuning of PSS problem is devised as nonlinear non differentiable optimization problem [10] which is
very hard to solve using traditionally differentiable
optimization algorithms. Different random exploration
techniques like Tabu Search (TS), Simulated Annealing (SA),
Genetic Algorithm (GA), Evolutionary Programming (EP)
has been used [11]-[15] to optimize the PSS parameters.
These techniques have gained acceptance for searching an
optimal solution in a problem space. GA is found to be very
satisfactory in searching global or near global optimal result
of the problem. But long run-time constraint limits the use of
GA in PSS design. Invasive Weed Optimization (IWO) and
Particle Swarm Optimization (PSO) were used to optimize the
PSS parameters [16]-[17]. Both algorithms have shown
satisfactory results but there are scopes to improve the
convergence time and the shifting of the critical mode to the
left of the s-plane. Newer algorithm can be used to obtain
better result.
I. INTRODUCTION
36
Ahmed, A., et al.: Elec. Engg., Instn. Engrs., Bangladesh, 39 (I&II), June & December 2013
lag generated between input and the output signals. The four
time constants T1to T4and the gain Kpss need to be chosen in an
optimal fashion for each generator to ensure optimal system
performance under various system loadings to ensure the
robustness of the technique.
X& = f ( X ,U )
(1)
B. Objective Function
For the design of optimal parameters for the PSSs, an Eigenvalue based objective function is considered as given in Eq.
(4).
X& = AX + B U
J =-min(real(Eigen-values)/abs(Eigen-values))
(2)
U pss ,i = K pss ,i
i
K pss ,i
sTwi
1 + sTwi
1 + sT1i
1 + sT3i
1 + sT2i
1 + sT4i
(4)
subject to,
(6)
(7)
(8)
(9)
(10)
(3)
U pss ,i
Here i and Upssi are the rotor speed deviation from the
synchronous speed and the power system stabilizer control
signal which is supplemented to the generator excitation
system. The washout block works as a high pass filter which
allows transient inputs and blocks the steady ones. From the
view of the washout function, the value of the time constant
Tw is generally not critical and may be in the range of 0.5 to
20 seconds [18]. In this work it is fixed at 10s. The phase
compensation block with time constants T1, T2, T3 and T4
supplies the required phase-lead to compensate for the phase
C. Population Structure
DE starts with a population of Np vectors of D dimensional
single-valued parameters as represented in Eq. (11).
(11)
xi , g
(12
2)
E. Mutation
on, DE mutatees the populatiion to producee a
Affter initializatio
poppulation of fou
ur trial vectorss. These four vectors from the
t
iniitial population
n are randomly
y sampled wheere one is chossen
as the target vector mentioned in the next sttage, and anoth
her
as the base vecctor. The diffference of thee remaining tw
wo
vecctors scaled by
y a factor is ad
dded to the basse vector to forrm
thee trial vector. Eq.
E 13 shows how
h
the processs of creating the
t
inttermediate vecttor is achieved.
Vi , g = xr 0, g + F . ( xr1, g xr 2,g
2 )
3)
(13
F. Crossover
In this stage, DE
E crosses each vector with a mutant vector to
forrm a second inttermediate pop
pulation as show
wn by Eq.(14).
0,1) Cr or j = jrand ]
Vj, i, g if [rand j(0
(14
4)
Ui,g =
Z j ,i, g otherwise
MULATION RE
ESULT
IV. SIM
As the location of PS
SSs in the netw
work is not amoong the focus
of the ppaper, generatoor and PSS nuumber are conssidered to be
equal hhere. Four tim
me constants foor each of the PSS having
two leaad-lag blocks aand a gain bloock will be tunned to attain
the opptimum value. So the num
mber of param
meters to be
optimizzed for this woork is 20.
G. Selection
Thhe selection off vectors to po
opulate the neext generation is
acccomplished by
y comparing each
e
vector ui,g
of the seco
ond
i
inttermediate pop
pulation Ug to its
i target vector xi,g from whiich
it iinherits parameters. The valu
ues of the vecttors are obtain
ned
usiing the function
n as illustrated
d in Eq. (15).
38
L7
Light
P
Q
17.67
-2.50
Nominal
N
P
Q
19..00
-3.00
Heavy
P
Q
20.00
-3.50
L9
9.67
9.6
67
10.50
-1.00
-1.25
-1.50
Table II:
I Optimal DE
E-Based PSS Paarameters
with objecctive function
Load
Gen
K
T1
T2
G1 32.3656
6 0.4114 0.4113
G2
50
1
0.010
00
G3 45.5009
9 0.6040
1
G4
50
0.0100
1
T3
0.0100
0.1655
0.3006
0.9025
T4
0.010
00
0.975
58
0.045
59
0.010
00
G1 32.3656
6 0.4114 0.4113
G2
50
1
0.010
00
G3 45.5009
9 0.6040
1
G4
50
0.0100
1
0.0100
0.1655
0.3006
0.9025
0.010
00
0.975
58
0.045
59 Nominal
0.010
00
G1
G2
G3
G4
0.4195
0.2065
0.6683
0.8092
0.446
62
0.658
89
0.692
24
0.015
51
33.3076
6
28.0287
7
37.5958
8
34.5677
7
0.4895
0.6746
0.5494
0.2440
0.6310
0.050
02
0.993
37
0.993
37
From T
Table II, it is seen that in ccase of light aand nominal
loadingg the parametters are identtical whereas for heavier
loadingg the optimizerr finds a differeent set of solutiion.
Light
Heavy
It is evvident that one complex set oof eigen value in each case
has possitive real part which would m
make the systeem dynamics
unstablle with oscillations of groowing magnituude. This is
becausee the positive real part introdduces negativee damping to
the sysstem and the ooverall system damping thuss gets worse.
The unnstable modes aare shown in boolded case in thhe table.
Nom
minal
0469
-90.0
-89.8
8503
-77.6
6029
-66.3
3374
-33.8
8824
-22.9
9197
-9.3
3958
-9.6
6185
-0.4582 7.5750i
-0.5139 7.3007i
0.1165 4.0628i
H
Heavy
-26.05 34.50i,0.755
-12.54 20.77i,0.603
-05.16 08.60i,0.600
-03.62 06.04i,0.599
-01.78 02.94i,0.605
-01.45 00.10i,14.50
-00.10 00.00i, 0.00
The m
maximum and minimum raanges for thee gains and
differennt time constannts for optimizzation purpose are set at [0
- 50] aand [0.01 - 11.0] respectiveely. The DE algorithm is
appliedd to find ouut the optimizzed value off the target
parameeters which meeet the objectiive in (4). Thhree different
loadingg conditions arre considered for L7 and L99 to examine
the apppropriateness oof the DE in finding an opptimal set of
parameeter. The loadinng scenario is provided in T
Table I. Final
values of the DE opttimized parameeters are providded in Table
II.
Load
Nominal
-27.25 48.81i,,0.558
-15.16 17.53i,0.864
-05.66 13.11i,0.431
-07.54 12.34i,0.611
-01.44 03.22i,0.447
-02.45 01.09i,2.247
-01.59 01.10i,1.445
Heavy
-90.0855
-89.8464
-78.4163
-66.6141
-33.6686
-22.2615
-9.8979
-9.2374
0.1549 4.1758i
-0..2629 7.7054i
-0..5634 7.2620i
39
Ahmed, A., et al.: Elec. Engg., Instn. Engrs., Bangladesh, 39 (I&II), June & December 2013
rotor angle variation (t)
-0.2
-0.3
delta(degree)
Light loading
Nominal loading
Heavy loading
-0.4
-0.5
-0.4
-0.6
-0.8
0
-0.6
0
50
100
150
200
Without PSS
DE based PSS
-0.2
d e lta (d e g re e )
-0.2
-0.3
Without PSS
DE based PSS
-0.2
-0.4
-0.6
-0.8
0
3
4
5
6
Time (sec)
Fig. 7 Rotor angle response in heavy loading for Gen2.
-0.4
-0.5
0
3
Time (sec)
250
iteration
-0.1
3
4
Time (sec)
Without PSS
DE based PSS
1.01
1.005
1
0.995
0
3
4
5
6
Time (sec)
Fig. 8 Rotor speed response in light loading for Gen2.
angular speed variation (t)
omega(rad/sec)
1.015
Without PSS
DE based PSS
1.01
1.005
1
0.995
0
3
Time (sec)
1.015
Without PSS
DE based PSS
1.01
1.005
1
0.995
0
3
Time (sec)
40
Ahmed, A., et al.: Elec. Engg., Instn. Engrs., Bangladesh, 39 (I&II), June & December 2013
for Gen 2.
stabilizer control signal
0
-20
0
2
3
Time (sec)
power(pu)
0
1
2
3
Time (sec)
20
0
1
2
3
Time (sec)
Upss
DE based PSS
0
-1
-2
0
3
4
5
6
Time (sec)
Fig. 14 Stabilizer control signal (Upss) in light loading
for Gen 2.
stabilizer control signal
Upss
DE based PSS
0
-1
-2
0
K. Comparative analysis
To have an understanding of the level of usefulness of DE based
tuning approach of the PSSs, a comparative analysis is performed
between this work and [17]. The comparisons related to the eigen
value based analysis is presented in Table V and that related to
the time domain based ones are shown in Table VI.
10
-10
0
3
Time (sec)
30
20
-20
0
Without PSS
DE based PSS
40
0
-0.5
0
0.5
power(pu)
DE based PSS
20
Upss
power(pu)
40
3
Time (sec)
41
Ahmed, A., et al.: Elec. Engg., Instn. Engrs., Bangladesh, 39 (I&II), June & December 2013
Light
Ref.
This
[17]
work
0.443
0.532
PSO
Nominal
Ref.
This
[17]
work
0.485
0.431
IWO
45
PSO
50
PSO
10
10
[6] M. A. Abido and Y. L. Abdel-Magid, A hybrid neurofuzzy power system stabilizer for multi-machine power
systems, IEEE Trans. Power Systems, vol. 13, pp.1323
1330, 1998.
Heavy
Ref.
This
[17]
work
0.5967
0.599
PSO
40
IWO
110
Light
Ref.
This
[17]
work
Nominal
Ref.
This
[17] work
Heavy
Ref.
This
[17]
work
0.12
0.11
0.1
0.1
0.08
0.15
1.2
IWO
1.2
2
PSO
1.2
1.1
IWO
1.0
V. CONCLUSIONS
In this paper the performance of the Differential Evolution
technique has been investigated for the design of Power System
Stabilizers in a multi-machine power system. An eigenvalue
value objective function is proposed for optimizing the PSS
performance. Eigenvalues and time domain basedanalyses are
carried out for three different loading conditions. The result
shows that DE can successfully optimize the PSS performance
for variable loading scenarios. Finally, a comparative analysis
reveals that there is no clear choice among the PSO, IWO or DE
based optimization technique for tuning the PSSs of the example
system. Use of hybrid optimization involving these three can be a
subject of future work.
REFERENCES
Ahmed, A., et al.: Elec. Engg., Instn. Engrs., Bangladesh, 39 (I&II), June & December 2013
rotor angle
rotor speed
Pm
mechanical input power
Pe
electrical output power
Eq
internal voltage behind xdi
Efd
equivalent excitation voltage
Te
electric torque
Tdo
time constant of excitation circuit
Tqo
regulator time constant
vref
reference voltage
v
terminal voltage
APPENDIX
Machine Model
= b(i- 1)
43