You are on page 1of 8

Design of a Fast Transient Stability Control Scheme in Power System

Zhihong Yu, Xiaoxin Zhou, Fellow, IEEE and Zhongxi Wu

AbstractThis paper describes the optimization of the settings for various emergency controls in an electrical power system, which is done by PSASP Lab of CEPRI. The aim is to regain a state of operating equilibrium when the power system is subjected to disturbances. Generator tripping with necessary load shedding is adopted as the main stabilizing control actions. A hybrid particle swarm approach is proposed to investigate the optimal combination of diversiform controls actuated after an event to maintain or restore the stability. In this way, the number of tripped generator units and shed loads can be minimized while stable operation can be maintained under extreme contingencies. Tests on the IEEE-39 New England power system and a large-scale practical Chinese utility, Southern Power Grid of China, are performed in the paper. The results indicate the feasibility of the proposed strategy. Index TermsTransient stability control, particle swarm optimization, generator tripping, load shedding.

I. INTRODUCTION ith the rapid increase in capitalized industrial loads, the electrical power is usually transmitted long distance to users by high capacity lines. Due to economical considerations, most of power systems operate near their stability limits. It is rather important for a power system to be stable after certain contingencies. Proper control actions should be taken to improve its transient stability further after the occurrence of large disturbances. At present, there are a variety of emergency controls that are used for transient stability enhancement of power systems, such as excitation control, generator tripping, fast valving, fast fault clearing, dynamic braking and load shedding. Among of them, generator tripping is regarded as one of the most effective approaches [1]. Generator tripping and load shedding are one-shot by nature and can be realized fast. In our study, generator tripping is selected as the main stabilizing control action. The load shedding is followed, if necessary. Once the controlled objects (generators and loads) are chosen, the higher-level controller will determine the appropriate amount of generation and load to regulate at each location, for arbitrary disturbance scenarios, and in the

presence of system uncertainty. Also, the disrupted amount of generation and load should be minimized so as to save operational costs and network losses. Traditionally, the deterministic decision-making is still utilized within system operation. Most decisions depend on the operators experience and numerous scenario simulations. But on the other hand, the power system has been shifting from a regulated system to a competitive and uncertain market environment, and the conditions under which power systems are operated have become more complex. This has led to operators to face more pressure, from economic imperatives in the marketplace to operate power systems with lower security margins, resulting in more frequent encounter with highly stressed conditions requiring operator fast decision. In response, a hybrid particle swarm optimization algorithm is developed. It is used to seek the optimal settings of various control actions automatically on a particular event while ensuring system transient stability. The amount of generation and load to be modulated are gained from snapshots of real operating conditions. Because of restrictive run-time requirements associated with online applications, together with the inherent increased computational demands of analyzing uncertainty, analysis speed is a critical issue in decision of transient stability control scheme. For this reason, Support Vector Machines (SVMs) technique is employed to enhance the computational efficiency of transient stability assessment. II. PARTICLE SWARM OPTIMIZATION ALGORITHM Particle Swarm Optimization (PSO) is a novel evolutionary computation technique, which was first introduced by J. Kennedy and R. C. Eberhart in 1995 [2]. It is motivated by the study on social behaviors of organisms, such as bird flocking and fish schooling. This technique has been verified to be robust for solving problems featuring non-linearity, non-differentiability, multiple optima, and high dimensionality through adaptation. PSO technique can generate high-quality solutions within shorter calculation time and have more stable convergence characteristic than other stochastic methods. Other attractive features of PSO include ease of implementation and the fact that no gradient information is required. The application of PSO in power systems may be as wide as possible. For example, [3] and [4] focused on the problems of reactive power and voltage stability control. [5] proposed a hybrid PSO for distribution state estimation. It has been found that the PSO can quickly find the high-quality optimal solution for many power system optimization problems.

This work was supported by the National Basic Research Program (973 Program) of China (2004CB217904). Zhihong Yu is with the PSASP Lab of CEPRI (Electric Power Research Institute of China ), Qinhe, Beijing,100085,China(e-mail: zhhyu@epri.ac.cn) Xiaoxin Zhou is with CEPRI, Qinhe, Beijing, 100085, China (e-mail: xxzhou@epri.ac.cn ) Zhongxi Wu is with the PSASP Lab of CEPRI, Qinhe, Beijing, 100085, China (e-mail: wzhongxi@epri.ac.cn)

1-4244-0493-2/06/$20.00 2006 IEEE.

As an optimization tool, PSO provides a population-based search procedure. Each individual called particle flies around in a multidimensional search space. It adjusts its velocity and position in the direction of its previous best position (solution or fitness) and the best previous position of any particle in the neighborhood. The former value is called pbest and the latter is called gbest. Thus, a PSO system can combine local search methods with global search methods, attempting to balance exploration and exploitation. A. Continuous PSO Suppose the ith particle is represented as xi=(xi1,xi2,,xid) in a d-dimensional space. The best previous position of the ith particle is recorded and represented as pbesti=(pbesti1, pbesti2,, pbestid).The rate of the position change (velocity) for particle i is represented as vi=(vi1,vi2,,vid). The index of the best particle among all the particles in the population is represented by the gbest. The particle-updating mechanism for particle flying (i.e., search process) can be formulated as: v(ik+1) =wv(ik) +a1rand(0,1)( pbest(ik) - x(i k) )+ a2Rand(0,1) (k) (k) (1) (gbest - xi ) x(ik+1) = x(ik) + v(Ik+1) (2)

will take on a binary value (0 or 1). III. IMPLEMENTATION OF THE PSO FOR TRANSIENT STABILITY
CONTROL SOLUTION

A. Fitness Function The fitness function provides a measure of how the individual performs in the problem domain. To the transient stability problem, the objective of emergency control is to maintain synchronism when subjected to a severe disturbance, while minimize the disrupted amount of generation and load. Generally, the absolute value of stability margin denotes the distance from the present operation point to the stability border. The larger the value, the more unstable or stable the system with respect to disturbances. So the fitness is defined as a function of stability margin and the tripping amount in our study. Equations (4) and (5) show the fitness values that PSO is to maximize. To an unstable case predetermined by a transient stability assessment (TSA) method, the primary aim is not to ensure the minimal usage of emergency controls, but to restore the stability. The fitness function can be described as: f=

abs (stability margin)

(4)

where a1 , a2 : acceleration constants; w : inertia weight; rand(0,1), Rand(0,1) : uniform random number between 0 and 1; v(ik) : velocity of individual i at iteration k, Vminv(ik) Vmax; x(ik) : position of individual i at iteration k; pbest(ik) : best position of individual i until iteration k; gbest(k): best position of the group of the swarm until iteration k. As described above, it can be seen that PSO is simple in concept, easy to implement, and computationally efficient. Furthermore, PSO provide a flexible and well-balanced mechanism to enhance and adapt to the global and local exploration abilities. B. Discrete Binary PSO [6] Because only whole generator units can be tripped in practical emergency controls, the amount of generator tripping cannot be adjusted in the same manner as the other controls, such as load shedding. The tripping proportion of each generator would be 0 or 1. For the reason, discrete binary PSO (BPSO) is used to determine the minimal number of tripping generator units here. In the binary discrete version, the trajectories are changes in the probability that a coordinate will take on binary values. The main difference between original PSO and the BPSO is that eq.(3) is replaced by eq.(2); If (rand ()<S(v
(k+1) i

After the power system is regulated and assessed to be stable, the main goal of the turning process is to minimize the required amount of generation tripping and load shedding. The corresponding fitness function can be written as:

f =abs (stability margin)+weight

(5)
NL j =1

MVA
i =1

NG

Gi

+ MVALj

MVA
i =1

NG

Gi

and

MVA
j =1

NL

Lj

refer to the summation amount of

generator and load reductions, where NG is the number of the predetermined tripping generators and NL is the number of the predetermined shedding loads of the studied system respectively. B. Transient Stability Control Algorithm Using PSO Before using PSO algorithm to solve the transient stability control optimization problem, the representation of the input particle must be defined. In our study, each particle xi is a vector with the dimension equal to (NG+NL). Each particle can be represented as Fig.1.

[1 0 1 0.85 0.6 0.9]


1 2 NG 1 2

NL

Fig.1. Representation of a particle in the population for the transient stability control problem solution

))

then x =1 else x = 0

(k+1) i (k+1) i

(3)

where S(v) is a sigmoid limiting transformation function (S(v)=1/(1+e-v)), and rand() is a quasi-random number selected from a uniform distribution in [0.0,1.0]. Each bit xid

The former NG dimensions represent the status of generators and the rest NL dimensions represent those of loads. Each generator gene denotes whether the generator is tripped. If the generator is tripped, the value will be 1. On the contrary, it will be 0. Each load dimension denotes the proportion of the load shedding. It is constrained to be within the range of 0
to 1.

If a power system is predicted to be unstable with time, hybrid PSO method will be used to calculate the minimal control amounts. The minimum is necessary for the system to maintain/restore equilibrium between system generation and load. Incorporating with the modifications above, the computational procedures of the proposed transient stability control algorithm using the hybrid PSO are shown in the following steps: 1. Determine the generators which should be tripped and the loads which should be shed. When a transmission line is opened due to a line fault, power flow on some tie lines will exceed their transfer limits. Excess generation of a local area may make all of the generators in this area unstable. Generators on the power output area and loads on the power input area will be selected as the preferential control objects. Each 2. Initialize the input particle swarm x={xi} N i=1 . individual xi is a (NG+NL)-dimensional vector. In this paper, to the first individual, its dimensions corresponding to the generators and loads, which were decided by step 1, are initialized to nonzero data, whereas the other bits are set as zero. To the rest individuals, their dimensions are initialized randomly. 3. In term of the value of each individual xi, set the disturb input file titled ST.S12 of PSASP, run transient stability assessment program, and calculate the evaluation value of each individual by transient stability assessment and the evaluation function f as given by eq.(4) or eq.(5) . (k+1) (k) 4. Update pbesti with x(ik+1) if f (x(ik+1) ) > f(pbesti ) and ( k +1) update gbest(k+1) with best pbesti in the population swarm. 5. Update the member velocity vi of each individual xi according to eq.(1) and modify the position of individual xi according to eq.(2) for load shedding and eq.(3) for
Decide which generators should be tripped and which loads should be shed

generator tripping. 6. if vid


(k+1) (k+1)

>vid , then vid


min

max

(k+1)

= vid

max

if vid >vid , then vid = vid 7. Go to step 3 until a maximum number of iterations criterion is reached. 8. Output the individual that generated the latest gbest control scheme, which represents the optimal combination amounts of generator tripping and load shed. Due to the dimension of search space decided by the step 1 is much larger than that of the last solution, much unacceptable computation time will be spent. Thus in our procedure, once an individual make the studied system become from unstable to stable, the search space will be redefined by its columns whose value unequal to zero. The remainder individuals will be reinitialized. The subsequent search will be directed towards areas in the new search space for which a match is more likely. Fig. 2 shows the procedures for generating feasible transient control strategy. IV. SPEED ENHANCEMENT BY SVM CLASSIFIERS Speed is a key issue for transient stability control in an online environment. In above procedure, the proposed control strategy is represented by a particle. A TSA method should be performed to test the effect of each particle at each iteration step and produce the best promising results. In general, detailed time-domain simulation for the complete strategy is the most stability assessment approach, but it is also the most time-consuming method for TSA. For an unstable case, transient stability assessment may be performed to obtain the local best tripping amount for each particle at each instruction. Even for a simple system, therefore, generation of the final optimal tripping amount need call TSA programs thousands of times. The time-consuming method will become a heavy burden for the strategy decision and makes it difficult for practical applications. To improve the speed of TSA, a quick prediction is imperative to judge the transient stability of a power system. So, the SVMs integrated with short-duration time-domain simulations are chosen to realize fast TSA in our approach. Support Vector Machines were invented by Vladimir Vapnik. They are a method for creating functions from a set of labeled training data. The function can be a classification function or the function can be a general regression function. For classification, SVMs operate by finding a hypersurface in the space of possible inputs. This hypersurface will attempt to split the positive examples from the negative examples. The split will be chosen to have the largest distance from the hypersurface to the nearest of the positive and negative examples. Intuitively, this makes the classification correct for testing data that is near, but not identical to the training data. The next paragraphs provide a brief introduction to the core concept of SVMs. More information about these can be found in [7] and [8]. Given a training data set of N Points {xi,yi}N i=1, where xi (xi n R ) is a n-dimensional input vector and yi (yiR) is the

(k+1)

min

Initial values

Run TSA program(time domain simulation or SVMs-based method, etc.)

Compute fitness value of each particle. If the corresponding case predetermined by TSA is unstable, use eq.(4); else use eq(5)

Search pbest i and gbest

Update the generator tripping and load shedding value

Satisfy stopping criteria?

Output result

Fig.2. The procedure for generating feasible transient control strategy.

associated output value of xi. The aim of SVMs is to construct a separating hyperplane as the decision surface, which divides the set of examples such that all the points with the same label are on the same side of the hyperplane. Meanwhile, the margin of decision boundaries is achieved to be maximal. In its basic form, a SVM classifies a pattern vector x into class y based on the support vectors xi and their corresponding classes yi as

w = wmax

wmax wmin
itermax

iter

(9)

y = sign( i yi K (x i , x) + b)
i =1

(6)

where wmax=0.9, wmin=0.4, iter is the current number of generations and the maximum iteration number itermax is set to 100. In addition, weight in eq.(5) is set to 100. To verify the feasibility and efficiency of the proposed transient stability control approach based on hybrid PSO, two power systems with different scale are chosen as the test systems, including a standard IEEE 39-bus system and a practical power system, China Southern Power Grid. A. IEEE 39-Bus New England Power System The well-known IEEE 39-bus system is shown in Fig.3. The network consists of 10 generator buses, 19 load buses and 34 branches.
8

The parameters i and b are determined by a linearly constrained quadratic programming (QP) problem like:
N 1 N min i i j yi y j K (x i , x j ) 2 i, j =1 i=1

(7)

37
10

s.t. i yi = 0
i =1

(8)

30 2

25

26

28

29 38
9

0 i C , i =1,K , N

K(,) is a symmetric positive-definite kernel function, i is lagrange multiplier and C>0 is the penalty parameter to control trade-off between training errors and complexity of the SVMs. Input features listed in Appendix have been chosen to describe the power system operating status. It should be mentioned that the dynamic features not only include features at or after the fault inception and fault clearing time, but also involve information at or after the control actions start and finish time. Only in this way, the selected features just can sufficiently represent the stability states with some emergency control strategies. The output variable has been chosen to be the two classes of interest: stable or unstable For a new, unseen operating point, first, system snapshots, such as generator rotor angles, speeds, and accelerating powers at special time (fault or disturbance initial/clearing time) are taken from the short-duration time-domain simulation (terminated at some step after the disturbances) results. Then, input features can be hereby calculated to form an input vector x and the operating state can be classified as either stable or unstable by counting eq.(6). TSA is a program totally separated from above PSO program. V. NUMERICAL EXAMPLES The hybrid PSO based approach was implemented using the C++ language and was executed on a 2.80GHz PC. Some parameters must be assigned before hybrid PSO is used to achieve the minimal emergency controls. According to the experiences of many experiments, the following parameters are selected as: The number of particles N=20, acceleration constant a1= a2=2.0, the limit of veloicty change vmax=1.0, vmin=1.0 and the procedure will be terminated if the number of iterations reaches 100. The inertia weight factor w is set by eq.(9)

18

17

27 24 16
6

35 21 19 23 22

15 4 14 5 6 7 8 9
3 2

39

12 11 31 10 32 13 34
5

20
4

33 36
7

Fig. 3.

IEEE 39-Bus New England power system

Suppose that the system operates under 120% of the base loading condition. A three-phase short-circuit permanent fault happened on line 11-10 near bus 10. Consequently, the line was opened to clear the fault 0.15s after the contingency. For the studied system is small, Power System Analysis Software Package (PSASP) was used to decide whether the system is stable with emergency controls. The traditional form of load control (shedding) is quite disruptive to consumers. Usually, load shedding is managed to avoid because of the discontent created by the clients. For the IEEE 39-bus power system, the first aim is to find a combination of dissimilar controls without any load shedding. If the search with no load shedding is unsuccessful, the algorithm essentially restarts a search with possible small load shedding that are effectively unnoticed by consumers related to the control combinations. The control scheme for IEEE-39 can be achieved by hybrid PSO algorithm. As a result, without load shedding, the case can be stabilized just with tripping generator unit at bus 39 at 0.2s after the event. The total tripping amount is 10.879 p.u. The convergence property of the hybrid PSO method is

shown in Fig.4 and Fig.5.


0.2 0 -0.2 -0.4 -0.6 -0.8 0 5 10 15 20 25 30 Iteration 35 40 45 50

Fig. 4. Convergence characteristics of IEEE-39 power system by the hybrid PSO method
25
Generator Tripping Amount

B. China Southern Power Grid The proposed hybrid PSO is applied to a practical large-scale power system, China Southern Power Grid (CSG). CSG is one of the two main power grids of China. Its service area covers Guangdong, Guangxi, Yunnan, Guizhou and Hainan Provinces, which possesses of an area about one million square kilometers and the total population of 220 millions. This Grid represents 39283 kilometers of transmission lines operating at 220 kV and greater with total transforming capacity 140.050 GVA from west to east by the end of 2004. So far, the total installed capacity has increased to 80.27GW (excluding that in Hong Kong and Macau) and the annual generating capacity has increased to 383.2TWh. The total society electricity consumption has increased to 389.1TWh and accounting for 17.9% of the total in China. The studied system in this paper has 2585 buses, 305 generator buses, 629 load buses, 1699 AC transmission lines and 3 500kV DC transmission lines.
35 32
1290/95/96, 1286/87/88 1243

Fitness Value

Tripping Amount(p.u.)

20
1255/6,1299

15

10

146 3600
14 7 /8

10
927,1061

893

82/6

29
29 30

4
94 2,9 86 20,

1
56

2
24 3

38
1147/9

12 15

145 53 57 54 58 143

1158/59,1200/03/80/81/82/58

11

33
11 75 18 /6,1 2/ 3 1239/48

36

37
1248/9

34
1226/7, 1250

18
170 , 29 6

9
850/1,903/5

17 2 17 7, 30

13
8 19

16 12

39

10

15

20

25 30 Iteration

35

40

45

50
219, 228

8
230 /1/ 3, 3 00

Fig. 5. Optimal tripping amount vs. number of iteration by the hybrid PSO method for IEEE-39 power system

15

19

Simulation results of this event are shown in Fig.6. Fig.6(a) shows the event without any control in 5 seconds. Fig.6(b) shows the same event stabilized after generator tripping. The Y-axis represents the angle difference between generator 30 and the other generators.
Generator Angles (deg.)
0 x 10
4

Fig.7. Schematic representation of the local network configuration of CSG power system

-1

-2 0

Time(s)

Fig. 6 (a). Swing curves of IEEE-39 unstable with no controls simulated.


Generator Angles (deg.)
200 100 0 - 100 - 200 - 300 0 1 2 3 4 5

Time(s)

Fig. 6 (b). Swing curves of IEEE-39 stable with generator tripping at bus 39 simulated.

Fig.7 presents a partial area chart of CSG, where each circle represents an area. Each directed line represents the tie lines in assemble and the corresponding power flow. Line 1-2 is a DC line. In this example, two three permanent fault on tie line 54 and 58 near area 10 is assumed to happen as shown in Fig.7. The lines are cleared within 0.2s by opening following lines: tie lines 53 and 57 between area 10 and 11 tie lines 56 between area 1 and 7 line 59 of area 1(including 31 branches) Simulation indicates that the loss of these lines couldnt stabilize the system. But we may find a combination of generator tripping and load shedding with hybrid PSO, which can maintain the system stability. Since the studied system is very large, it would be time-consuming and inefficient in searching all possible combination of generator and load reduction. Buses to be cut after the faults should be decided first. Based on the power flow analysis, generator buses in areas (11, 33, 34, 37) and load buses in area 2 are selected to construct the PSO search space. Then, each PSO particle

traverses the search space so as to find the global minimal generator tripping and load shedding requirements. All active controls are assumed to occur 0.22s after the faults. Table I presents the results of detail tripping amounts.
TABLE I AMOUNT OF TRIPPING GENERATORS AND SHEDDING LOADS

Fig.10 shows dynamic responding of CSG with or without controls after the faults. Fig.10 (a) shows the event with no control simulated for 20 seconds and Fig. 10 (b) shows the same event stabilized after generator tripping and load shedding. It should be pointed out that the transient stability criterion used for this case is set by CSGs operators. Loss of synchronism across the network is determined when the maximum angle difference between any two generators exceeds 500 degrees.
14000 12000 10000 8000 6000 4000 2000 0 - 2000
Generator Angles (deg.)

10
Time(s)

12

14

16

18

20

Fig. 10 (a).
500
Generator Angles (deg.)

Swing curves of CSG unstable with no controls simulated;

The convergence property of the hybrid PSO method is shown in Fig.8 and Fig.9. It can be seen that the proposed hybrid PSO method can converge at the optimal solutions in a short time.
16 14 12

- 500 0

10
Time(s)

12

14

16

18

20

Fig. 10 (b). shedding.

Swing curves of CSG stable with generator tripping and load

10 8 6 4 2 0 0 10 20 30 40 50 60 70 80 90 100

Iteration

Fig.8.
70

Convergence property by mosified PSO for CSGs case

Generator Tripping Amount

60

Load Shedding Amount

Tripping Amount(p.u.)

50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100

Each particle, which represents a stability control strategy, is verified through a TSA method. For a large-scale power system, such as CSG, the time needed for TSA will be excessive, if it is accomplished through a full-duration time-domain simulation. To enhance the computational efficiency, short-duration time-domain simulation and SVM classifier are employed here. In the study, the numerical simulation is for 1s dynamic process and provide the input features for the following SVMs classifier. The information can also be derived from the synchronized phasor measurements in the future. The required SVMs classifier is designed by 1000 labeled or known training patterns in advance. For a particle, per iteration averagely spends 34s in achieving the fitness value by time-domain simulation, while less than 2s by the hybrid framework. Assuming that the population number is 15 and the iteration number is 20, the complete time-domain simulation will spend 34 s*15*20=170min to obtain the final result, while the time wiill be shorten to 2s*15*20=10min by the hybrid framework. VI. CONCLUSIONS Our study focused on using the hybrid PSO algorithm to derive combinations of emergency controls to stabilize unstable transient events in simulation. Two different PSO indices are used to derive combinations of various controls to stabilize strongly unstable transient events in simulation. The first serves for those unstable cases, and is based on the change of stability degree with some controls. The other is presented as a summation of all tripping amount, employed when the system restores to a normal state. Continual PSO and discrete binary PSO are used for

Fitness Value

Iteration

Fig.9. Optimal Tripping Amount vs. number of iteration by the hybrid PSO method for CSG power system

generator tipping and load shedding problems respectively. Global or near global optimum solution for optimal combination of generator and load reduction can be ascertained within a reasonable time. It is optimal for the enhancement of transient stability under different fault conditions. Meanwhile, the significant economic loss of the excessive control can also be avoided. The selection of the initial tripping generators and shedding loads are crucial to the final results. Proper definition decides the convergence of the PSO algorithm and the reliability of the control amount. In our approach, power flow analysis is used to set the control objects, which can be chosen by the sensitivity analysis and by the practical engineering experience at future times. The power system stability can be maintained by a scheme of generator tripping and load shedding. Further work will focus on other one-shot controls, such as HVDC fast power change and reactive power switching, etc. VII. APPENDIX According to many references and simulation results, features used in this paper are listed as follows: 1. Initial relative rotor angle of the machine with the maximal rotor acceleration rates 2. Relative rotor angle of the machine with the maximal rotor kinetic energy at the control action time 3. Difference of rotor angle between lead machine and rear machine, where lead machine is the generator with the maximal initial rotor acceleration rate and rear machine is the generator with the minimal initial rotor acceleration rate 4. Difference between the maximal and minimal rotor angle at the control action time 5. The maximal relative rotor angle at the control action time 6. Difference between the relative rotor angular velocity at the control action time and the initial relative rotor angular velocity 7.

differential coefficient of rotor acceleration rate at the control action time 18. The maximal value of all the rotor kinetic energies at the control action time 19. The rotor kinetic energies of the machine with the maximal rotor angle at the control action time 20. Total system "energy adjustment", x20 = (Pai di ) ,
i =1

where Pai is the acceleration power of the ith machine, and

di is the relative rotor angle at the control action time


1 NG x = M i Pdi 21 21. M 0 i =1 22. Mean value of all the rotor kinetic energies at the control action time 23. Mean value of all the initial mechanical power of generators 24. Standard deviation of all the initial mechanical power of generators 25. Mean value of all the initial rotor acceleration power 26. Standard deviation of all the initial rotor acceleration power 27. Mean value of all the initial relative rotor acceleration power 28. Standard deviation of all the initial relative rotor acceleration power 29. Total system rotor kinetic energy at the control action time 30. Difference between the maximal and minimal rotor kinetic energy at the control action time 31. Difference between the maximal and minimal differential coefficient of rotor kinetic energy at the control action time
32. x32 = ci 0i 33. x33 = max { ci 0i }
i
i =1 NG

x7 = ci , where NG is the number of generators and


i =1

NG

ci is the rotor angular velocity at the control action time

34. x34 = ( (Pmi Pei ) M i )


i =1

NG

8. x8 =max ci 9. Maximal initial relative rotor acceleration rate 10. Minimal initial relative rotor acceleration rate 11. Mean value of all the initial rotor acceleration rates 12. Standard deviation of all the initial rotor acceleration rates 13. Mean value of all the initial relative rotor acceleration rates 14. Standard deviation of all the initial relative rotor acceleration rates 15. Difference of initial rotor acceleration rates between lead machine and rear machine 16. Difference between the maximal and minimal rotor acceleration rate at the control action time 17. Difference between the maximal and minimal

tPC tPC

35. x35 = max { (Pmi Pei ) M i }


i

tpc refer to the instant after the control action Features at fault clear time are defined as those at control action time. VIII. REFERENCE [1] J. Machowski, Power System Dynamics and Stability. New York: Wiley, 1997. [2] R. Eberhart and J. Kennedy and, "Particle swarm optimization," Proceedings of IEEE International Conference on Neural Networks (ICNN95), pp.1942-1948, 1995. [3] H. Yoshida et al., "A particle swarm optimization for reactive power and voltage control considering voltage

[4]

[5]

[6]

[7] [8]

security assessment," IEEE Trans. Power Syst., vol. 15, no. 4, pp. 12321239, Nov. 2000. L.L. Lai, T.Y. Nieh, D. Vujatovic, Y.N. Ma, Y.P. Lu, Y.W. Yang and H. Braun, "Swarm intelligence for optimal reactive power dispatch," Conference Proceedings, 2005 IEEE/PES Transmission and Distribution Conference & Exhibition: Asia and Pacific, China, August 2005, IEEE Catalog Number: 05CH37678C S. Naka et al, "A hybrid particle swarm optimization for distribution state estimation," IEEE Trans. Power Syst., vol. 18, no. 1, pp. 6068, Feb. 2003. J. Kennedy and R. Eberhart, "A Discrete Binary Version of the Particle Swarm Optimization," Proceedings of IEEE International Conference on Systems, Man, and Cybernetics (SMC'97), pp.4104-4109, 1997. V.Vapnik, Statistical Learning Theory. New York: John Wiley and Sons, Inc, 1998. Christopher J.C.Burges, "A tutorial on support vector machines for pattern recognition," Data Mining and Knowledge Discovery, 2(2), pp. 121-167, Apr. 1988. IX. BIOGRAPHIES

Zhihong Yu was born in Xinjiang, China, in 1975. She received her B.Sc. and M.Sc. degrees from Xinjiang University, China, in 1997, 2000 respectively. She received the Ph.D degree in electrical engineering from Harbin Institute of Technology in 2004. She is currently working in CEPRI. Her current research includes power system analysis and data mining. Xiaoxin Zhou Academician of Chinese Academy of Science, IEEE Fellow. His current research interests include power system analysis and control, the real time simulation of power system, and FACTS. Zhongxi Wu Professor in PSASP Lab of CEPRI.

You might also like