IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 1099 Generalized Integer Linear Programming Formulation for Optimal PMU Placement Bei Gou, Member, IEEE Abstract—Based on the integer linear programming formulation proposed for optimal PMU placement, this paper presents a generalized integer linear programming formulation for cases including redundant PMU placement, full observability and incomplete observability. Due to accurate voltage phasor measurement and current phasor measurements provided by PUM units, the accuracy, redundancy and thus the robustness of state estimation will be enhanced with the integration of PMU measurements. The problem of optimal placement of PMU for the redundant PMU placement, full observability and incomplete observability analysis needs to be studied, for various purposes and considerations. The proposed modeling approach by the author in another paper, which models PMU placement as an integer linear programming problem, is extended and generalized to satisfy different needs. Cases with and without zero injection measurements are considered. Simulation results on different power systems show that the proposed algorithm can be used in practice. Index Terms—Integer linear programming, observability analysis, phasor measurement units. I. INTRODUCTION P HASOR measurement units (PMUs) become more and more attractive to power engineers because they can provide synchronized measurements of real-time phasors of voltage and currents [1]. As the sole system monitor, state estimator plays an important role in the security of power system operations. Optimal placement of PMUs in power systems to enhance state estimation is a problem needed to be solved. Several algorithms and approaches have been published in the literature. An algorithm which finds the minimal set of PMU placement needed for power system state estimation has been developed in [2] and [3] where the graph theory and the simulated annealing method have been used to achieve the goal. In [4] a strategic PMU placement algorithm is developed to improve the bad data processing capability of state estimation by taking advantage of the PMU technology. Techniques for identifying placement sites for phasor measurement units in a power system based on incomplete observability are presented in [1] where simulated annealing method is used to solve the pragmatic communication–constrained PMU placement problem. In [5] a special tailored nondominated sorting genetic algorithm is developed for the PMU placement problem. The authors in [5] developed an optimal placement algorithm for PMUs by using integer programming. However, the proposed integer programming becomes a nonlinear integer programming under the existence of conventional power flow or power injection measurements. In this paper optimal PMU placement problem is re-studied for the cases of redundant PMU placement, full observability and incomplete observability. Based on the author’s newly proposed integer linear programming algorithm in [7], the PMU placement problem is re-studied and a generalized integer linear programming formulation is presented in this paper. Numerical tests show that this formulation is efficient and able to obtain the identical results as those in previous publications. II. PRELIMINARY RESULTS OF FULL OBSERVABILITY Unlike traditional measurement units, the PMU is able to measure the voltage phasor of the installed bus and the current phasors of all the lines connecting to the bus. That is, a PMU can make the installed bus and its neighboring buses observable. The objective of placing PMUs in power systems is to determine a minimal set of PMUs such that the whole system is observable. A. Without Conventional Measurements A PMU, different from traditional meters, is able to measure the voltage phasor of the installed bus and the current phasors of all the lines connecting to this bus. That is to say, a PMU can make the installed bus and its neighboring buses observable. Therefore the placement of PMUs becomes a problem that finds a minimal set of PMUs such that a bus must be “reached” at least once by the set of the PMUs. This gives us an idea to define a matrix . are defined as follows: The entries in if if and are connected otherwise. The optimal placement of PMUs is formulated as follows [7]: where is the PMU placement variable and Manuscript received October 30, 2007; revised March 17, 2008. Paper no. TPWRS-00794-2007. The author is with the Energy System Research Center, University of Texas at Arlington, Arlington, TX 76019 USA (e-mail: bgou@uta.edu). Digital Object Identifier 10.1109/TPWRS.2008.926475 0885-8950/$25.00 © 2008 IEEE .. . 1100 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 B. With Conventional Measurements Let us define a vector . The element of indicates the number of times for bus “reached” by PMUs, where is the th row of and is the th element of . For detail discussion, the following three cases need to be analyzed. 1) If a power flow measurement is on line , then the following needs to be held: which means one bus voltage can be solved from this measurement and the other needs to be covered by PMU. 2) Let us assume that bus connects buses , and . Suppose that an injection measurement is at bus , then the following inequality needs to be held: 3) The above power flow measurements and the injection measurements are associated. According to the approaches introduced in 1) and 2), we have the following two inequalities: Fig. 1. Example for illustration. C. Example I In order to clearly explain the above formulation of integer linear programming for the case of full observability, we use the example in [6] to illustrate it. Suppose that a branch flow measurement be on line 2–3 and an injection measurement at bus 2. Bus 1, 2, 3, 6 and 7 are associated to these two conventional measurements. According to the definition given above, we have and In order to satisfy , the first inequality needs to be subscribed from the second inequality corresponding to the injection measurement and consider that its right hand side needs to be reduced by one due to the injection , the second inequality be. comes Therefore, the inequality for this case is If bus is not associated to any conventional measurements, then the corresponding constraint of the minimization problem . in Section II-A is still kept When considering the conventional measurements, the optimal placement of PMUs can be formulated as a problem of integer linear programming as follows: Buses 4 and 5 are not associated to these two conventional measurements. The two inequality constraints corresponding to these two conventional measurements are Therefore, we have Branch Measurement Injection Measurement Then matrix can be written as follows: The permutation matrix where the matrix ; the matrix is a permutation matrix and is the number of buses not associated to conventional measurements; the column vector is defined in [7]. The details of forming these matrices are given in the following examples. is GOU: GENERALIZED INTEGER LINEAR PROGRAMMING FORMULATION FOR OPTIMAL PMU PLACEMENT 1101 Therefore, the integer linear programming for this PMU optimal placement can be written as Fig. 2. Simple example. two terminals of any branch, is larger than 1. Therefore, the depth-of-one unobservability PMU placement can be formulated as the following integer linear programming problem: The optimal solution of this integer linear programming is , which means the PMU needs to be placed as buses 2 and 5. This example shows that the conventional measurements do not affect the decision of optimal placement because of the system configuration and the locations of conventional measurements. III. PMU PLACEMENTS FOR INCOMPLETE OBSERVABILITY ANALYSIS A novel concept, incomplete observability, has been introduced in [1], where incomplete observability refers to the network condition in which the number and location of the PMUs are insufficient to determine the complete set of bus voltages (the state). Also, the depth of unobservability is a measure of the distance of an unobserved bus from its observed neighbors [1]. Based on the depth of unobservability, the concepts depth-of-one unobservability, depth-of-two unobservability, and so on are also defined in [1]. The details of the definitions can be found in [1]. Based on the idea proposed in [7], formulation of the incomplete observability PMU placement problem can be studied using the similar modeling approach and solution method, without and with the considerations of zero injections. In this paper we will only study the cases of depth-of-one unobservability and depth-of-two unobservability with the understanding that depth of higher degrees can be dealt with in the same manner. where , is the total number of is the PMU placement variable, and is the branches, branch-to-node incident matrix. 2) Depth-of-Two Unobservability: Depth-of-two unobservability means that the at most two unobservable buses can be connected in the system, which means that the sum of , corresponding to any three connecting buses, is larger than 1. Therefore, this problem can be formulated as follows: where , is the total number of posis the PMU sible combinations of three connecting buses, placement variable, and is the matrix, each row of which corresponds to three connecting buses and contains all of the possible combinations of three connecting buses. B. With Zero Injection Measurements If a bus has neither generation nor load, the sum of flows on all the associated branches to the bus is zero. This equality normally corresponds to a zero injection measurement and can be used in state estimation. 1) Depth-of-One Unobservability: Let us use Fig. 2 as an example. Zero injection measurement at bus implies that A. Without Zero Injection Measurements 1) Depth-of-One Unobservability: Depth-of-one unobservability is defined as a situation in which all of the neighboring buses of any unobservable bus must be observable. The definition implies that it is impossible for two unobservable buses to connect together, which provides us the way to develop the integer linear programming for this case. To obtain the formulation, a matrix needs to be deare given in Section II-A. fined. The details of If we define a vector , the element of indicates the number of times that bus can be is the th row of . calculated by PMUs, where So that the depth-of-one unobservability can be modeled as a set of linear inequalities, the sum of , corresponding to the The equalities and as well as imply that Since and is the vector of binary variables is always true. Therefore, it is not necessary that this be listed in the formulation, as illustrated below. 1102 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 Also, since branch 1, 2, 3 and 4 are associated to the zero is injection measurement, matrix The final integer linear programming for this example is Fig. 3. Example for incomplete observability. Based on the problem , we have where is the matrix that keeps the branches that are not associated with zero injection measurements and removes the branches that are associated with zero injection measurements. . Others are the same as in 2) Depth-of-Two Unobservability: Let us use the same example as in the previous section. Depth-of-two unobservability implies that we have the following inequalities: and The solution is . So if a PMU is placed at bus 3, then we can have depth-of-one unobservability for this system. 2) Depth-of-Two Unobservability : For this system, the total number of all possible three connecting buses is 15. Since a defined zero injection measurement is at bus 2, the matrix in Section III-B can be formed as follows: which imply when the zero injection measurement at bus is considered. Therefore, the problem can be formulated as follows: Matrix is formed as follows: where is the matrix that keeps the combinations that are not associated with zero injection measurements and removes the branches that are associated with zero injection measurements. . Others are the same as in C. Example II Let us use the same system as in Fig. 1. Here we only show the case with zero injection measurement. We assume that a zero injection measurement is at bus 2 (check Fig. 3 for details). is 1) Depth-of-One Unobservability: For this case, the same as in example I. The branch-to-node incident matrix is given as Branch Branch Branch Branch Branch Branch Branch Branch Then the integer linear programming can be written as GOU: GENERALIZED INTEGER LINEAR PROGRAMMING FORMULATION FOR OPTIMAL PMU PLACEMENT The solution of this problem is , which means a PMU needs to be placed at bus 4 to reach the depth-of-two unobservability. 1103 TABLE I NUMBER OF PMU PLACEMENT FOR INCOMPLETE OBSERVABILITY WITHOUT ZERO INJECTIONS IV. PMU PLACEMENT FOR REDUNDANT MEASUREMENT PLACEMENT The formulation of optimal PMU placement under the cases of full observability and incomplete observability has been presented in Sections II and III. However, the measurements redundancy is critical for bad data detection and identification in state estimation. Different levels of measurements redundancy can be requested for optimal PMU placement. As in previous sections, the two cases with and without zero injection measurements, are considered. The requirements of the measurement redundancy can be formulated through the column vector in the right hand side of the linear inequality constraints in the integer linear programming formulation for the cases with and without zero injection measurements. The measurement redundancy can be equal at all buses or different at different buses, which completely depends on the requirements of the system and the budget. V. GENERALIZED INTEGER LINEAR PROGRAMMING FORMULATION The above linear programming formulation can be further generalized for the cases of redundant PMU placement, full observability and incomplete observability. The generalized form is given as follows: TABLE II NUMBER OF PMU PLACEMENTS FOR INCOMPLETE OBSERVABILITY WITH ZERO INJECTIONS In Table II, we consider the zero injection measurements and not nonzero injection measurements. The number of zero injection measurements is listed in Table II, respectively, for three power systems. It should be noted that the number of required PMUs of the proposed integer linear programming algorithm for different systems is identical to that in [1]. However, the locations of required PMUs are different, which means that the problem of PMU placement with incomplete observability has multiple solutions. VII. CONCLUSION where matrix it the transformation matrix which transforms into matrix the requirements of the cases of redundant PMU placement, full observability or incomplete observability; the indicates the redundancy requirements for column vector all buses. and take different values according to different cases, based on the above results in previous sections. VI. SIMULATION RESULTS We have tested the proposed formulation on IEEE 14-, 30-, and 57-bus systems. We use the binary integer programming of MatLab to solve this problem. The proposed integer linear programming algorithm has been tested on different systems. The results without and with zero injection measurement are displayed in Tables I and II. In Table I, we do not consider any zero and nonzero injections; that means, we only consider the placement of pure phasor measurements provided by PMUs. This paper proposes a generalized integer linear programming formulation for optimal PMU placement under different cases of redundant PMU placement, full observability and incomplete observability. This generalized formulation, considering the situations with and without zero injection measurements, shows that the problem of optimal PMU placement can be modeled linearly and can be solved by integer linear programming. The generalized formulation paves an efficient way for future research in PMU placement and related topics. Simulation results show that the proposed algorithm is computationally efficient and can be used in practice. REFERENCES [1] R. F. Nuqui and A. G. Phadke, “Phasor measurement unit placement techniques for complete and incomplete observability,” IEEE Trans. Power Del., vol. 20, no. 4, pp. 2381–2388, Oct. 2005. [2] L. Mili, T. Baldwin, and R. Adapa, “Phasor measurement placement for voltage stability analysis of power systems,” in Proc. 29th Conf. Decision and Control, Honolulu, HI, Dec. 1990. [3] T. L. Baldwin, L. Mili, M. B. Boisen, and R. Adapa, “Power system observability with minimal phasor measurement placement,” IEEE Trans. Power Syst., vol. 8, no. 2, pp. 707–715, May 1993. 1104 IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 23, NO. 3, AUGUST 2008 [4] J. Chen and A. Abur, “Placement of PMUs to enable bad data detection in state estimation,” IEEE Trans. Power Syst, vol. 21, no. 4, pp. 1608–1615, Nov. 2006. [5] B. Milosevic and M. Begovic, “Nondominated sorting genetic algorithm for optimal phasor measurement placement,” IEEE Trans. Power Syst., vol. 18, no. 1, pp. 69–75, Feb. 2003. [6] B. Xu and A. Abur, “Observability analysis and measurement placement for systems with PMUs,” in Proc. 2004 IEEE Power Eng. Soc. Conf. Expo., Oct. 10–13, 2004, vol. 2, pp. 943–946. [7] B. Gou, “Optimal placement of PMUs by integer linear programming,” IEEE Trans. Power Syst., vol. 23, no. 3, pp. 1525–1526, Aug. 2008. Bei Gou (S’97–M’00) received the B.S. degree in electrical engineering from North China University of Electric Power, Beijing, China, in 1990, the M.S. degree from Shanghai JiaoTong University, Shanghai, China, in 1993, and the Ph.D. degree from Texas A&M University, College Station, in 2000. From 1993 to 1996, he taught in the Department of Electric Power Engineering at Shanghai JiaoTong University. He worked as a research assistant at Texas A&M University beginning in 1997. Subsequently, he worked at ABB Energy Information Systems, Santa Clara, CA, for two years, and at ISO New England for one year as a senior analyst. He is currently an assistant professor in the Energy Systems Research Center (ESRC) at the University of Texas at Arlington. His main interests are power system state estimation, power market operations, power quality, power system reliability, and distributed generators.