IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO.

4, NOVEMBER 2011

2371

An Eigenstructure-Based Performance Index and Its Application to Control Design for Damping Inter-Area Oscillations in Power Systems
D. P. Ke, C. Y. Chung, Senior Member, IEEE, and Yusheng Xue, Member, IEEE
AbstractAn eigenstructure-based performance index is proposed in this paper to measure the dynamic performance of the system as well as control efforts. Calculation of this index is based on eigenstructure of the closed loop system and the design parameters; it does not rely on control structures. Therefore, this index can be applied for solving structurally constrained control problems. A tuning scheme based on this index is proposed for coordinating power system stabilizers (PSSs) and supplementary damping controllers (SDCs) for exible AC transmission systems (FACTS) devices to damp inter-area oscillations of systems and to optimize their control efforts under multiple operating conditions. Both PSSs and SDCs utilize control structures as a low order single-input-single-output phase lead-lag compensator. Wide-area signals are employed to upgrade their effectiveness in damping inter-area oscillations. Time delays caused by usage of wide-area signals are also considered in the tuning scheme. Results of simulation on a four-machine two-area system and the New England and New York interconnected system show that the proposed index is effective in measuring dynamic performance of the system and the coordinatedly tuned PSSs and SDCs based on this index can robustly damp inter-area oscillations of systems with optimized control efforts. Index TermsCoordination, damping control, eigenstructure, inter-area oscillation, structural constraint.

I. INTRODUCTION

OWER system stabilizers (PSSs) have been widely employed to provide additional damping for inter-area oscillations in power systems [1][3]. It is also recognized that supplementary damping controllers (SDCs) for exible AC transmission systems (FACTS) devices have great potential in damping inter-area oscillations [4]. Generally it is expected that as these damping controllers (PSSs and SDCs) work in a coordinated manner to provide adequate additional damping to
Manuscript received September 16, 2010; revised January 16, 2011; accepted February 22, 2011. Date of publication March 28, 2011; date of current version October 21, 2011. This work was supported in part by Research Grants Council of Hong Kong (PolyU 5154/08E) and in part by the Department of Electrical Engineering of The Hong Kong Polytechnic University. Paper no. TPWRS-00748-2010. D. P. Ke and C. Y. Chung are with the Computational Intelligence Applications Research Laboratory (CIARLab), Department of Electrical Engineering, The Hong Kong Polytechnic University, Hong Kong (e-mail: 08901306r@polyu.edu.hk; eecychun@polyu.edu.hk). Y. Xue is with State Grid Electric Power Research Institute, State Grid Corporation of China, Nanjing 210003, China (e-mail: xueyusheng@sgepri.sgcc. com.cn). Color versions of one or more of the gures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identier 10.1109/TPWRS.2011.2123119

inter-area oscillations, their control efforts should be optimal [5], [6]. Moreover these controllers should be robust for different operating conditions or network congurations [7], [8]. Besides, structural constraints are usually imposed on damping controllers: they must use dynamic output feedback control schemes because it is difcult to implement state feedback control in power systems; they should be low order and with a simple structure familiar to engineers so that implementation and the subsequent tuning are easy [7][9]. Moreover, although a decentralized implementation scheme using only local signals as inputs is more practical for damping controllers, they can be congured for quasi-decentralized implementation when wide-area signals are employed as inputs to enhance their effectiveness in damping inter-area oscillations [9]. However time delays that occur because of use of wide-area signals should be considered in the design. Several methods have been proposed for the design of damping controllers. Approaches based on robust control theories and linear matrix inequalities (LMI) have been applied for damping controller design to deal with uncertainties of operating conditions [5], [8], [10][12]. However they cannot be applied for design of structurally constrained controllers. In [8], for designing structurally constrained SDCs for coordinated functioning, the LMI-based regional pole placement problem is converted into the iteratively solved bilinear matrix inequalities (BMI) problem. Nevertheless, this method assumes that input matrices of state space equations of the controllers are known. Methods that directly optimize eigenvalues of the closed loop system can readily be applied to structurally constrained controllers while considering multiple operating conditions [7], [13][16]. However, system dynamics in time domain are not only related to eigenvalues, but also associated with eigenvectors [17]. Moreover, control effort cannot be explicitly optimized in these methods. Therefore, one salient merit of optimal control [6], [9] is that the control process, together with the system dynamics objective, can be explicitly considered in the cost function. Unfortunately standard optimal control cannot be applied to structurally constrained controllers although it can perform well for state feedback controllers [12]. The method applied in [9], [18], and [19] tries to solve this problem by assuming that poles of synthesized controllers are known. However this assumption is only applicable for some certain control structures and the method cannot take multiple operating conditions into consideration. In this paper, an eigenstructure-based performance index is proposed to measure the system dynamic performance as well as the control efforts. Computation of this index does not rely on

0885-8950/$26.00 2011 IEEE

2372

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 4, NOVEMBER 2011

the control structures because it is only related to the eigenstructure of the closed loop system and the design parameters. Minimizing this index can improve system dynamics and optimize the control efforts. Therefore, by utilizing this index to construct objective functions, structurally constrained controllers can be designed for optimization of control. Based on this idea, a tuning method is proposed for coordination of structurally constrained PSSs and SDCs as stated above, to damp inter-area oscillations as well as to optimize the control efforts under multiple operating conditions. This paper is organized as follows. In Section II, the proposed index is introduced. The closed loop power system model used to calculate the index is synthesized in Section III. In Section IV, the proposed tuning scheme for coordination of PSSs and SDCs is introduced. Simulation results are reported in Section V. Conclusions are presented in Section VI. II. PROPOSED EIGENSTRUCTURE-BASED PERFORMANCE INDEX Irrespective of the structure adopted by damping controllers, the synthesized linear model of a closed loop power system around an operating point can generally be described as follows: (1) (2) (3) where and are the state vector and state matrix, respectively, for the th operating condition taken into consideration; is the output vector representing system dynamics objective; is the output vector of damping controllers; and are output matrices. and For the th operating condition, a quadratic performance measurement (cost function) is dened as (4) where and are diagonal matrices with positive entries on their respective diagonals; H is the conis the initial value of . As jugate transpose operator; and variables denoting relative power angles of generators are often selected as components of , minimization of (4) will suppress power angle oscillations of the system and also optimize control efforts of damping controllers. Substituting (2) and (3) into (4), can be rewritten as (5) where is a Hermite matrix dened as (6) In power system dynamic analysis, state matrix can be similarly diagonalized [20], [21]. Thus the time domain solution of (1) can be derived as (7) where and spectively, of are right and left eigenvector matrices, reis a diagonal matrix dened as follows: (8)

are eigenvalues of . where are with If the system is stable, i.e., all eigenvalues of negative real parts, then (5) can be calculated based on (7), as follows:

(9) is termed as cost matrix for the th operating condiwhere tion and is dened as follows: (10) Here denotes dot production and the following denition: is a Hermite matrix with

. . .

. . .

. . .

. . .

(11)

where is the conjugate operator. is positive for any given . It is seen from (4) that is a positive Therefore, from (9) and (10), it is inferred that denite matrix. Consequently (9) can be further decomposed as follows:

(12) (13) where are eigenvalues of and are real are the corresponding positive numbers; onto the orthogonal coordinates values when projecting . It is noted that basis formed by the right eigenvectors of are only related to the eigenstructure of the , and ) and design parameters ( , system ( ) for the th operating condition. and denotes the initial disturbed deviation from the Actually, operating point and it cannot be determined in the control deare also unsign process [5], which means that is a direct time domain indetermined. Therefore, though dicative of the dynamic performance of the system and has been utilized in the state feedback optimal control [6], it is generally directly for designing the structurally ineffective to use cannot constrained controllers because the undetermined be dealt with in such cases. Nevertheless, according to (12), for the undetermined is to reduce a way to reduce , and vice versa. Therefore, a new performance index is proposed in this paper as follows: (14) Obviously this index is independent of , and it is not equivalent to but it too can measure the performance of system

KE et al.: EIGENSTRUCTURE-BASED PERFORMANCE INDEX AND ITS APPLICATION TO CONTROL DESIGN

2373

C. Time Delay Approximation Time delays are approximated by using the second order Pade formula [9]: (17) where is the time delay vector. The state space equations describing the dynamics of time delays are obtained as follows:
Fig. 1. Overall system structure.

dynamics as well as the control efforts. Furthermore, irrespective of control structures, derivation of this index can be just based on the synthesized closed loop system model. is the function of parameters of damping It is clear that controllers. The system dynamics as well as control efforts can . be optimized by adjusting these parameters to minimize , the linear model of the closed However in order to calculate loop system has to be constructed; this is introduced in the next section. III. MODELING OF THE CLOSED LOOP SYSTEM A. System Structure The structure of the overall system in which PSSs and SDCs work in a coordinated manner to damp inter-area oscillations is illustrated in Fig. 1. Both PSSs and SDCs are assumed to be a classical phase lead-lag compensator. They are implemented in a quasi-decentralized manner, and wide-area signals are employed to enhance their effectiveness in damping inter-area oscillations. Possible time delays are approximately considered in the design. The modeling of each part in Fig. 1 and synthesis of the closed loop system model are presented in the following subsections. B. Reduced-Order Open Loop Power System Model The order of a practical power system model is usually high, but most internal dynamics of the model can be neglected without causing considerable inaccuracy in describing the input-output relationship since frequencies of the concerned system dynamics are between 0.2 and 0.8 Hz (inter-area oscillations) [12]. Thus, the low-order approximated models obtained via model reduction can be employed to facilitate the subsequent design process [6], [9], [11]. There are a variety of model reduction techniques suitable for reducing the order of practical power system models [22], [23]. Therefore, to reduce the time and to accelerate the tuning required for computation of process, the Schur balanced model truncation algorithm [24] is applied in this paper to obtain the following reduced-order open loop power system model: (15) (16) where and are the state vector and state matrix, respectively, of the reduced-order system for the th operating condition; is the output vector for inputs of damping controllers; is the input matrix; and are output matrices.

(18) (19) where is the state vector of time delays; is the input vector of damping controllers; , and are state matrix, input matrix, output matrix, and feed-forward matrix, respectively. By incorporating (18) and (19) into (15) and (16), the following linear model is obtained: (20) (21) with the following matrix denition:

D. Modeling of Controllers Suppose there are damping controllers (specically ) and the th controller has parameters in Fig. 1, and . These controllers can be modeled as (22) (23) and are output and input, respectively, of the th where is the state vector; and the following matrices are controller; dened:

Then state space equations of the synthesized controller are expressed as (24) (25)

2374

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 4, NOVEMBER 2011

where

and are lower and upper limits, respectively, of ; calculation of these limits is presented in the following. Suppose there is a phase lead-lag block shown as follows: (29) If the phase of this block reaches maximum (or minimum) of at frequency , then time constants and can be determined by (30) (31) When the maximum and minimum compensated phases provided by the block and frequencies of inter-area modes are and can be calculated given, upper and lower limits of depending on (30) and (31), respectively. Upper and lower limits of the gain of the phase compensator can be calculated by residue analysis [8]. B. Solving the Optimization Problem The optimization problem (26)(28) is a standard constrained nonlinear programming problem (NLP) solved in this paper by sequential quadratic programming (SQP), a highly effective and matured method for the NLP [25]. Initial values of controller parameters used as a starting point for the SQP are given by the conventional sequential tuning method, which is also employed for comparison with the proposed tuning scheme [16]. Firstly, the compensated phase supplied by a controller to an inter-area mode is derived by residue analysis in the nominal operating condition [12]. The gain of this controller is then increased gradually to enhance damping of the mode while considering the control effort of the controller as well as side effects on other modes. Each controller is tuned sequentially while the other already tuned controllers are online. To depict the solving process more conveniently, (26)(28) are expressed in a more general and compact form as follows: (32) (33) (32)(33) are solved by an iteration process based on the Lagrangian function, constructed as follows: (34) is the Lagrangian multiplier vector for . At the where beginning of the th iteration, the controller parameter vector and the positive denite Hessian matrix (which is initially an identity matrix and is updated iteratively to nally converge to the real Hessian matrix of the Lagrangian function) are available. Then, the following steps are executed [25]: S.1) Formulate and solve the following convex quadratic programming (QP) subproblem: (35) (36)

The linear model (1)(3) can then be constructed by incorporating (24) and (25) into (20) and (21). The corresponding matrix relationships are obtained as follows:

Consequently the proposed index can be calculated from the synthesized linear model of the closed loop system. A tuning scheme based on this index is proposed in the next section to simultaneously adjust parameters of PSSs and SDCs to damp inter-area oscillations and to optimize their control efforts under multiple operating conditions. IV. PROCEDURE OF CONTROLLER TUNING A. Tuning Problem Formulation Choosing variables to form should take into account two points: one is that inter-area modes should be sufciently obso that minimizing will result in suppression served in of inter-area oscillations; and another is that if the damping of a mode (i.e., a local mode) deteriorates dramatically after controller tuning due to its poor visibility in , variables that signicantly participate in this mode should be included in reconso that it can be considered in controller tuning struction of and thus appropriately damped. To ensure robustness of the controllers, typical multiple operating conditions are considered in the design. In this paper, , and of PSSs and SDCs are asparameters and washout sumed to be adjustable, while lter constant are preset and remain xed during the tuning time constant process. Therefore, an optimization-based tuning scheme for coordination of PSSs and SDCs is proposed as follows: (26) (27) (28) where is the number of operating conditions; is the weight is the damping ratio of of the th operating condition; is a real positive number (2% in this paper) to ensure some small signal stability margin while the objective of system damping control is achieved by minimization of (26); and is the parameter vector with the following denition:

KE et al.: EIGENSTRUCTURE-BASED PERFORMANCE INDEX AND ITS APPLICATION TO CONTROL DESIGN

2375

where is the search direction vector of parameters at the th iteration; and T are Hamilton and transpose operators, respectively. S.2) Based on , the controller parameter vector at the next iteration is calculated as follows: (37) where is the optimal step length along the search direction and it can be determined by minimizing the following merit function: (38) where is the penalty parameter vector. S.3) If the stopping criteria are satised, terminate the iteration process; otherwise, go to the next step. S.4) The new Hessian matrix is obtained by using the quasi-Newton method, as follows: (39) (40) (41) (42) (43) S.5) Set and go to the next iteration. , and

Fig. 2. Control structure of Statcom.

Based on the initial guesses, a trial process is then performed , and until acceptable controller tuning reto adjust sults for all operating conditions are derived [9]. The adjustment of design parameters is according to the following heuristic prinmay increase ciples: enhancing corresponding entries of damping of the system, while controller outputs would be signif; increasing could icantly constrained by larger entries of give more control priority to the th operating condition. V. EXAMPLES OF APPLICATIONS OF THE PROPOSED INDEX AND THE TUNING METHOD A. Four-Machine Two-Area System The classic four-machine two-area system is employed to demonstrate the proposed performance index and its application in simultaneous tuning of PSSs and SDCs. The diagram and parameters of this system can be found in [21]. A static compensator (Statcom) is installed at Bus 8 to maintain its voltage. The steady voltage at Bus 8 will rise to nearly 1.0 p.u. with reactive power support from the Statcom. The Statcom is modeled as a current injection always kept in perpendicular with bus voltage so that there is only reactive power exchange between the grid and the Statcom (Fig. 2) [26]. The voltage reg= 100 and ulator of the Statcom is an inert block, with = 0.005. The unit of time constants used in this paper is seconds, unless otherwise specied. The loads are modeled as a combination of constant impedances (50%) and induction motors (50%). The 47-order open loop system, therefore, has 24 state variables of generators (power angle, angular speed, d-axis and q-axis transient voltages, and d-axis and q-axis sub-transient voltages), 16 state variables of excitation systems, one state variable of the Statcom and six state variables of induction motors (angular speed, and d-axis and q-axis voltages behind transient reactance). Operating conditions considered for this system are as in Table I, where the rst ve are used for design while the last one is applied for validation of robustness of the controllers. The loads remain xed for all operating condition. Two local modes in this system are well damped. However, an inter-area mode with frequencies at about 0.65 Hz residing between Area 1 and Area 2 is poorly damped. Hence, an SDC is equipped in the Statcom, together with a PSS installed in Generator 4, to provide additional damping for inter-area oscillation. PSS is installed in Generator 4 on the basis of analysis of participation factors. Selection of wide-area feedback signals is based on residue analysis. Large residues of a systems input-output pair, with respect to a mode, indicate that this mode can be effectively controlled by the input-output pair using closed loop feedback control [12]. Accordingly, it is found that active power in line 10-9 is the most effective input signal for both PSS and SDC in damping

C. Selection of Design Parameters

Firstly, the appropriate initial guesses for , and are can be rewritten in an determined. From (6), (10), and (14), alternative expression, as follows:

(44) where and are th and th diagonal and , respectively; is the -dimensional entries of is computed from vector and its th component when setting , the rest of entries of are is the -dimensional vector and zeros and th component is computed from when its , the rest of entries of are zeros setting and . Depending on the initial values of controller and can be calculated. It is expected parameters, that all additive terms on the right-hand side of (44) can be so that all weighted in the same order of magnitude in components of and can be equivalently considered in and are optimization. Hence, the initial guesses for determined by simply setting these additive terms equal. With can be calculated. Thus, the initial these initial guesses, guesses for are derived by assuming that all additive terms in objective function (26) are equal as well.

2376

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 4, NOVEMBER 2011

TABLE I OPERATING CONDITIONS FOR THE FOUR-MACHINE TWO-AREA SYSTEM

Fig. 4. Searching process of SQP.

TABLE II CONTROLLER PARAMETERS

Fig. 3. Open loop frequency response of output of SDC to input of SDC.

inter-area oscillation and, therefore, it is chosen as the control input for these two controllers. The communication latency will be around 20 ms for sending this signal to the remote SDC site through a dedicated ber-optic communication channel [27], [28]. Moreover, since the time required for phasor measurement (about three 60-Hz cycles or 50 ms [27]) and signal processing is also considered, the total delay of 80 ms in feedback signal for the SDC is used in this design. Furthermore, the time delay in feedback signal for the PSS is assumed to be zero. The lter time constant and the washout time constant are set to 0.01 and 10, respectively, for both controllers. Since the inter-area mode is mainly dominated by the relative motions of generators between the two areas, output vectors used to form the cost function are dened as follows: (45) (46) , and are power angles of Generators 1, 2, 3, where and are outputs of the SDC and and 4, respectively; and the PSS, respectively. The open loop frequency response of output of the SDC to input of the SDC is illustrated in Fig. 3. A 15-order reduced model is obtained through model reduction of the original 47-order system and time delay is approximated by the Pade formula. Model reduction for each operating condition takes about 0.082 s; all time consumption tests in this paper are conducted in a desktop computer with 2.66-GHz CPU and 2G RAM. It is clearly seen that the approximated model can be employed to accurately represent the full model within the frequency range of interest. Lower and upper boundaries LB and UB of controller param, eters are given in Table II. Here parameters are corresponding to the SDC, while , and

and are for the PSS. Initial as well as tuned values of these parameters are also shown in the table. The searching process for solution of (26)(28) is illustrated in Fig. 4 and the time cost for nding this solution is about 1.74 s. It is clearly seen that the SQP is quite efcient in solving the proposed controller tuning problem. Eigenvalues of cost matrices are calculated when the PSS and the SDC are simultaneously tuned by the proposed method and are sequentially tuned by the conventional method. The rst four eigenvalues for each operating condition considered in the design are depicted in Fig. 5. The rst is the dominant one; it is much larger than the remaining. It is clear that this eigenvalue is obviously reduced when the proposed controllers are installed, compared to when the system is equipped with sequentially tuned controllers. Moreover, from (14), it is known is dened as the sum of eigenvalues of the cost matrix, that which means the proposed controllers will result in smaller than that given by sequentially tuned controllers. Because is capable of indicating the systems dynamic performance, it is naturally inferred that simultaneously tuned controllers will lead to a better dynamic performance of the system than sequentially tuned controllers. This is veried by computation of the closed loop system eigenvalues and the time domain simulations shown in the following. The inter-area mode eigenvalues of the closed loop system for all operating conditions are presented in Table III. The comparison shows that the proposed controllers truly provide more additional damping to the inter-area mode than sequentially tuned controllers not only for operating conditions used for design but also for the condition included in validation; damping of the

KE et al.: EIGENSTRUCTURE-BASED PERFORMANCE INDEX AND ITS APPLICATION TO CONTROL DESIGN

2377

Fig. 6. Power angle oscillations (solid line: proposed; dot line: conventional; dash line: no controller).

Fig. 5. Eigenvalues of cost matrices.

TABLE III EIGENVALUES OF INTER-AREA MODE

Fig. 7. Control signals (solid line: proposed; dot line: conventional).

inter-area mode is enhanced considerably by the proposed controllers. Because inter-area oscillations are adequately observed in can effectively measure the performance of these interarea oscillation dynamics according to the inference drawn in suppresses inter-area oscillaSection II. Minimization of tions and thus inevitably increases damping of the inter-area mode. Accordingly, though the proposed index is directly related to eigenvalues of the cost matrix, rather than the closed loop system, lowering it can actually enhance damping of closed loop system modes strongly associated with . This is veried as reduction of accords well with increase of damping of the inter-area mode shown above. A three-phase short circuit fault occurs at Bus 6 when the system is in operating condition 5 and the fault is cleared 50 ms later. The power angle oscillations between Generators 2 and 4 are shown in Fig. 6. It is seen that inter-area oscillation decays quite fast when the system is equipped with simultaneously tuned controllers, compared to the marginally stable system without controllers and the system with sequentially tuned controllers. Outputs of PSS and SDC are also optimized when they are coordinately designed by the proposed method (Fig. 7). The and , limits for outputs of PSS and SDC are set to respectively. Therefore, together with the above eigen-analysis, these results indicate that the proposed index is an effective measurement of performance of system dynamics as well as control efforts. Furthermore the proposed simultaneous tuning method based on this index for coordination of PSS and SDC can well damp inter-area oscillation under multiple operating conditions.

Fig. 8. New England and New York interconnected system.

B. New England and New York Interconnected System The equivalent (16-machine ve-area) model of New England and New York interconnected networks (Fig. 8) is employed to demonstrate performance of the proposed controller design including multiple inter-area modes. This system is a modied version of [12]: the quite large mechanical damping coefcients in the original model have been removed from some generators, and local PSSs are then installed in some generators to damp their local mode oscillations. A thyristor controlled series capacitor (TCSC) is installed in transmission line 50-18 to compensate 50% of its reactance in the steady state. The allowable TCSC dynamic compensation is from 10% to 90% of line reactance. The dynamic model of the TCSC is shown in Fig. 9 is chosen to be 10 ms. A com[11], where time constant bination of constant impedances (50%) and induction motors (50%) is employed to model the loads. Thus, there are 96 state variables of generators (the same six-order model as in the rst example), 64 state variables of excitation systems, 18 state variables of local PSSs, one state variable of TCSC, and 57 state

2378

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 4, NOVEMBER 2011

TABLE V CONTROLLER PARAMETERS

Fig. 9. Dynamic model of TCSC.

TABLE IV OPERATING CONDITIONS FOR THE 16-MACHINE FIVE-AREA SYSTEM

variables of induction motors in this system. By changing network congurations and transmitting different levels of power from area A1 to A2, nine typical operating conditions (Table IV) are considered in this study. Eigen-analysis shows that for all operating conditions, there are two quite poorly damped inter-area modes in this system: M1 and M2. M1 with frequency at about 0.65 Hz is dominated by the oscillation between generators in areas A1 and A2, while M2 with frequency at about 0.32 Hz depicts oscillations of generators in areas A1 and A2 with respect to generators in the rest of the system. To provide additional damping to the two inter-area modes, an SDC equipped in TCSC is designed. Meanwhile since Generator 13 greatly participates in both inter-area modes, a PSS is installed in Generator 13 to work in coordination with the SDC for damping control. Furthermore according to residue analysis, active power in transmission line 1317 is selected as control input for both PSS and SDC. Time delay is assumed to be 80 ms for transmitting the signal to the remote SDC. Participation factor analysis shows that Generators 13, 5, and 6 greatly participate in M1, while Generators 14, 15, and 13 greatly participate in M2. Therefore output vectors are formed as follows: (47) (48) , and are power angles of Generators where and are outputs of 5, 6, 13, 14, and 15, respectively; and SDC and PSS, respectively. The rst six operating conditions in Table IV are used for design, while the last three are applied for validation. It costs about 3.29 s to reduce the 236-order open loop power system model to a 28-order reduced model which can lead to satisfactory results of controller tuning by the proposed method. The lower and upper boundaries, the initial values, and the tuned values of controller parameters are shown in Table V. , and are for the SDC, Here parameters , and are for the PSS. Specically, it while

Fig. 10. Eigenvalues of cost matrices.

takes about 15 iterations and 6.18 s for the SQP method to converge. It is found that the controller parameter searching process for this much higher order system is still as efcient as that in the rst example, although the time cost (3.29 s) for model reduction in this system is relatively larger, compared to that (0.082 s) in the rst small system. This means that the proposed tuning scheme can be applied to large scale power systems due to the efcient reduced-order model-based controller parameter searching process. Eigenvalues of cost matrices for the rst six operating conditions are illustrated in Fig. 10. It is obvious that they are reduced by optimization. According to the verication shown in the rst example, the proposed controllers perform better than the sequentially tuned controllers. This is indeed conrmed again by facts shown in Table VI, that the proposed controllers provide more damping to both inter-area modes than the sequentially tuned controllers. The two inter-area modes are sufciently damped for all operating conditions when the proposed controllers are installed. A three-phase short circuit fault occurs at Bus 60 when the system is in operating condition 1. The fault is cleared 100 ms later by tripping one of tie-lines between Bus 60 and 61. Oscillations of relative power angles and active power in key tie-lines are depicted in Figs. 11 and 12, respectively. The compensation percentage provided by the TCSC and the output of the PSS during the dynamics are illustrated in Fig. 13. It is clear that simultaneously tuned controllers outperform sequentially tuned controllers, and the two inter-area oscillations are well damped

KE et al.: EIGENSTRUCTURE-BASED PERFORMANCE INDEX AND ITS APPLICATION TO CONTROL DESIGN

2379

TABLE VI EIGENVALUES OF INTER-AREA MODE

Fig. 13. Dynamics of controllers (solid line: proposed; dot line: conventional).

VI. CONCLUSION A novel eigenstructure-based performance index is proposed to solve structurally constrained control problems. This index can measure performance of system dynamics as well as control efforts. Calculation of this index has no bearing on control structures; it is based only on eigenstructure of the closed loop system and the design parameters. This index has been applied to a proposed optimization-based tuning scheme used for coordination of structurally constrained PSSs and SDCs to damp interarea oscillations and to optimize their control efforts under multiple operating conditions. Applications of the proposed control design method on a four-machine two-area system and the New England and New York interconnected system have been demonstrated. Results show that the proposed index is an effective way of measuring systems dynamic performance and reducing this index can improve the system dynamics. Structurally constrained PSSs and SDCs simultaneously tuned by the proposed method can effectively damp inter-area oscillations and their control efforts are also optimized. Future works will include applying the proposed damping controller tuning scheme to real large scale power networks in Mainland China, and also extending application of the proposed index and control design method to other control systems although this paper focuses only on application of the proposed method to PSSs and SDCs in power systems.

Fig. 11. Power angle oscillations (solid line: proposed; dot line: conventional; dash line: no controller).

ACKNOWLEDGMENT The authors would like to thank the anonymous reviewers for their valuable comments and suggestions to this research.

REFERENCES
Fig. 12. Oscillations of active power in key tie-lines (solid line: proposed; dot line: conventional; dash line: no controller). [1] I. Kamwa, R. Grondin, and G. Trudel, IEEE PSS2B versus PSS4B: The limits of performance of modern power system stabilizers, IEEE Trans. Power Syst., vol. 20, no. 2, pp. 903915, May 2005. [2] C. Y. Chung, K. W. Wang, C. T. Tse, and N. Riu, PSS design by probabilistic sensitivity indices, IEEE Trans. Power Syst., vol. 17, no. 3, pp. 688693, Aug. 2002. [3] H. G. Far, H. Banakar, P. Li, C. Luo, and B. T. Ooi, Damping interarea oscillations by multiple model selectivity method, IEEE Trans. Power Syst., vol. 24, no. 2, pp. 766775, May 2009.

by coordinately tuned PSS and SDC with optimum control outputs.

2380

IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 26, NO. 4, NOVEMBER 2011

[4] U. P. Mhaskar and A. M. Kulkarni, Power oscillation damping using FACTS devices: Modal controllability, observability in local signals, and location of transfer function zeros, IEEE Trans. Power Syst., vol. 21, no. 1, pp. 285294, Feb. 2006. [5] R. A. Ramos, A. C. P. Martins, and N. G. Bretas, An improved methodology for the design of power system damping controllers, IEEE Trans. Power Syst., vol. 20, no. 4, pp. 19381945, Nov. 2005. [6] A. C. Zolotas, B. Chaudhuri, I. M. Jaimoukha, and P. Korba, A study on LQG/LTR control for damping inter-area oscillations in power systems, IEEE Trans. Control Syst. Technol., vol. 15, no. 1, pp. 151160, Jan. 2007. [7] B. Chaudhuri, S. Ray, and R. Majumder, Robust low-order controller design for multi-modal power oscillation damping using exible AC transmission system devices, IET Gen. Transm. Distrib., vol. 3, no. 5, pp. 448459, 2009. [8] R. V. de Oliveira, R. Kuiava, R. A. Ramos, and N. G. Bretas, Automatic tuning method for the design of supplementary damping controllers for exible alternating current transmission system devices, IET Gen. Transm. Distrib., vol. 3, no. 10, pp. 919929, 2009. [9] D. Dotta, A. S. Silva, and I. C. Decker, Wide-area measurementsbased two-level control design considering signal transmission delay, IEEE Trans. Power Syst., vol. 24, no. 1, pp. 208216, Feb. 2009. [10] R. A. Ramos, L. F. C. Alberto, and N. G. Bretas, A new methodology for the coordinated design of robust decentralized power system damping controllers, IEEE Trans. Power Syst., vol. 19, no. 1, pp. 444454, Feb. 2004. [11] B. Chaudhuri and B. C. Pal, Robust design of multiple swing modes employing global stabilizing signals with a TCSC, IEEE Trans. Power Syst., vol. 19, no. 1, pp. 499506, Feb. 2004. [12] B. Pal and B. Chaudhuri, Robust Control in Power Systems. New York: Springer, 2005. [13] Z. Wang, C. Y. Chung, K. P. Wong, C. T. Tse, and K. W. Wang, Robust power system stabilizer design under multioperating conditions using differential evolution, IET Gen. Transm. Distrib., vol. 2, no. 5, pp. 690700, 2008. [14] I. Kamwa, G. Trudel, and L. Gerin-Lajoie, Robust design and coordination of multiple damping controllers using nonlinear constrained optimization, IEEE Trans. Power Syst., vol. 15, no. 3, pp. 10841092, Aug. 2000. [15] D. Z. Fang, S. Q. Yuan, Y. J. Wang, and T. S. Chung, Coordinated parameter design of STATCOM stabilizer and PSS using MSSA algorithm, IET Gen. Transm. Distrib., vol. 1, no. 4, pp. 670678, 2007. [16] L. J. Cai and I. Erlich, Simultaneous coordinated tuning of PSS and FACTS damping controllers in large power systems, IEEE Trans. Power Syst., vol. 20, no. 1, pp. 294300, Feb. 2005. [17] N. Kshatriya, U. D. Annakkage, F. M. Hughes, and A. M. Gole, Optimized partial eigenstructure assignment-based design of a combined PSS and active damping controller for a DFIG, IEEE Trans. Power Syst., vol. 25, no. 2, pp. 866876, May 2010. [18] A. J. A. S. Costa, F. D. Freitas, and H. E. Pena, Power system stabilizer design via structurally constrained optimal control, Elect. Power Syst. Res., vol. 33, no. 1, pp. 3340, Apr. 1995. [19] A. J. A. S. Costa, F. D. Freitas, and A. S. e Silva, Design of decentralized controllers for large power systems considering sparsity, IEEE Trans. Power Syst., vol. 12, no. 1, pp. 144152, Feb. 1997. [20] J. J. Sanchez, V. Vittal, M. J. Gibbard, A. R. Messina, D. J. Vowles, S. Liu, and U. D. Annakkage, Inclusion of higher order terms for smallsignal (modal) analysis: Committee report-task force on assessing the need to include high order terms for small-signal (modal) analysis, IEEE Trans. Power Syst., vol. 20, no. 4, pp. 18861904, Nov. 2005. [21] P. Kundar, Power System Stability and Control. New York: McGrawHill, 1994. [22] F. D. Freitas, J. Rommes, and N. Martins, Gramian-based reduction method applied to large sparse power system descriptor models, IEEE Trans. Power Syst., vol. 23, no. 3, pp. 12581270, Aug. 2008. [23] I. Kamwa, G. Trudel, and L. Gerin-Lajoie, Low-order black-box models for control system design in large power systems, IEEE Trans. Power Syst., vol. 11, no. 1, pp. 303311, Feb. 1996.

[24] M. G. Safonov and R. Y. Chiang, A Schur method for balanced model reduction, IEEE Trans. Autom. Control, vol. AC-34, no. 7, pp. 729733, Jul. 1989. [25] J. Nocedal and S. J. Wright, Numerical Optimization. New York: Springer, 2006. [26] M. H. Haque, Improvement of rst swing stability limit by utilizing full benet of shunt FACTS devices, IEEE Trans. Power Syst., vol. 19, no. 4, pp. 18941902, Nov. 2004. [27] C. W. Taylor, D. C. Erickson, K. E. Martin, R. E. Wilson, and V. Venkatasubramanian, WACS-Wide-area stability and voltage control system: R&D and online demonstration, Proc. IEEE, vol. 93, no. 5, pp. 892906, May 2005. [28] J. W. Stahlhut, T. J. Browne, G. T. Heydt, and V. Vittal, Latency viewed as a stochastic process and its impact on wide area power system control signals, IEEE Trans. Power Syst., vol. 23, no. 1, pp. 8491, Feb. 2008.

D. P. Ke received the B.S. and M.S. degrees in electrical engineering from Huazhong University of Science and Technology, Wuhan, China, in 2005 and 2007, respectively. Currently, he is pursuing the Ph.D. degree in electrical engineering at The Hong Kong Polytechnic University, Hong Kong, China. His research interests are in power system dynamics and control.

C. Y. Chung (M01SM07) received the B.Eng. degree (with First Class Honors) and the Ph.D. degree in electrical engineering from The Hong Kong Polytechnic University, Hong Kong, China, in 1995 and 1999, respectively. After his Ph.D. graduation, he worked in the Electrical Engineering Department at the University of Alberta, Edmonton, AB, Canada, and Powertech Labs, Inc., Surrey, BC, Canada. Currently, he is the Convenor of the Power Systems Research Group and an Associate Professor in the Department of Electrical Engineering, The Hong Kong Polytechnic University. His research interests include power system stability/control, planning and operation, computational intelligence applications, and power markets. Dr. Chung was the Chairman of the IEEE Hong Kong Joint Chapter of PES/IAS/PELS/IES in 20072009. During his tenure as chairman, the Joint Chapter received the 2008 IEEE PES Outstanding Small Chapter Award and the 2009 IEEE IAS Outstanding Small Joint Chapter Award. He was the Technical Chairman of IET APSCOM2009 International Conference, and Honorary Secretary of IEEE DRPT2004 International Conference and IEEE IAS 2005 Annual Meeting. Currently, he is the Chairman of the IEEE Hong Kong Section.

Yusheng Xue (M87) is an academician of the Chinese Academy of Engineering (CAE). He is also the Honorary President of State Grid Electric Power Research Institute (SGEPRI or NARI), China. He holds the positions of Adjunct Professor in many universities in China. Mr. Xue is a member of the PSCC Council and has been the Chairman of Technical Committee of Chinese National Committee of CIGRE since 2005.