Monocriteria and Multicriteria Optimization of Network Configuration in Distribution Systems R.C. BERREDO E.S. CRUZ Department of Engineering and Coordination of Distribution Expansion Planning Minas Gerais State Energy Company Ave. Barbacena, 1200, 30190-131 - Belo Horizonte - MG BRAZIL http://www.cemig.com.br P.Ya. EKEL M.F.D. JUNGES M.M. GONTIJO J.G. PEREIRA Jr. Graduate Program in Electrical Engineering Pontifical Catholic University of Minas Gerais Ave. Dom Jose Gaspar, 500, 30535-610 - Belo Horizonte - MG BRAZIL http://www.ppgee.pucminas.br V.A. POPOV Graduate Program in Electrical Engineering Federal University of Santa Maria Camobi, University Campos, 97105-900 – Santa Maria - RS BRAZIL http://www.ufsm.br/ppgee Abstract: - This paper presents results related to the project aimed at developing models, methods, and a computing system associated with problems of optimizing network configuration in distribution systems. Three lines of improving the validity and efficiency in solving these problems are considered. The first line is directed at taking into account a power system reaction while minimizing power and energy losses. The second line is related to formulating and solving the problems within the framework of multicriteria optimization models to consider and to optimize diverse indices reflecting reliability, service quality, and economic feasibility of energy supply. The third line is associated with taking into account the uncertainty of initial data. Adequate and computationally effective approaches to considering the factors indicated above are considered in the paper. Key-Words: - Distribution Systems, Network Reconfiguration, Power System Reaction, Multicriteria Optimization, Bellman-Zadeh Approach, Uncertainty Factor, Fuzzy Preference Relations. 1 Introduction The problems of optimizing network configuration (network reconfiguration) are associated with altering distribution system topological structures by changing the state of their switches (in other words, by changing locations of their disconnections). These problems are traditionally solved in long- and shortterm planning, dispatcher operation, and can be applied in design studies. An increased interest to the network reconfiguration problems is associated with wide automation of distribution systems whose switches are remotely monitored and controlled, that permits one to solve these problems as the on-line, real time problems. Many works have been dedicated to their solution on the basis of diverse approaches: search methods, branch and bound type techniques, heuristic procedures, expert systems, genetic algorithms, neural networks, etc. (for example, [1-4]). These works "compete" in aspiration for obtaining "more optimal" solutions. However, this aspiration, considering that combination of the information uncertainty (which is not considered in the papers [14]) and relative stability of optimal solutions produces decision uncertainty regions [5,8], is not convincing. At the same time, the results of [1-4] as well as other works in this field do not permit one to consider a power system reaction when optimizing network configuration in distribution systems. The operating conditions of distribution and power systems are related: changes of distribution network configuration lead to redistributing substations loads and, consequently, to changing losses in the power system. The lack of considering the change of power system losses may result not only to deterioration in network reconfiguration efficiency, but may result to the negative effect as well. It demands to minimize total losses in the distribution and power systems. This statement serves for real increasing the adequacy of models and, as a result, factual efficiency of solutions, which can be obtained on their basis. Moreover, the works [1-4] as well as other works in this field are directed to solving the problems on a monocriteria basis. At the same time, the problems of network reconfiguration are multicriteria in nature because have an impact on power and energy losses, reliability, and service quality. Taking the above into account, the work reported in this paper is dedicated to solving problems of monocriteria and multicriteria optimization of network configuration in distribution systems with taking into account the power system reaction and the uncertainty factor. 2 CONSIDERATION OF POWER SYSTEM REACTION Direct taking into account the power system reaction is hampered because of a large volume of information reflecting parameters and operating modes of distribution and power systems. Considering this, the paper is devoted to building functional equivalents to evaluate power system reaction at any step of distribution system optimization (under transference or an attempt of transference of any disconnection location for an arbitrary distribution network loop). The active power losses in the power system for an arbitrary step of bus load curves J p ,1 J q ,1 ... ... J p ,l J q ,l J J p jJ q ... j ... J p,m J q,m ... ... J p , n J q , n may be calculated in the following form [9]: P 3(J tp RJ p J tq RJ q ) 10 3 , J J p jJ q J p ,1 J q ,1 ... ... J p ,l J p ,lm J q ,l J q ,lm ... ... (3) j J p ,m J p ,lm J q ,m J q ,lm ... ... J p ,n J q ,n produced by transferring a location of disconnection of the distribution network loop connecting the buses l and m. This redistribution leads to an increment of power losses which may be estimated as follows: (Plm ) 3(J pt RJ p J qt RJ q J tp RJ p J tq RJ q ) 10 3 . (4) However, taking into account the properties of the matrices R and the fact that in the process of optimization we have to calculate the increments of losses ( Plm ) not for all combinations of the power system buses l (1 l n) and m (1 m n) (it is enough to consider pairs of substation buses (feed buses), which are connected by the distribution system), this approach is not sufficiently effective. An alternative approach is based on constructing functional equivalents to calculate ( Plm ) only for combinations of feed buses connected by the distribution system. In particular, the transformation of (4) with taking into account (1) and (3) leads to the following expression: (Plm ) 3{(I p2,lm I q2,lm )(Rll Rmm 2Rlm ) 2[(J p,lm J p,l J q,lm J q,l )( Rll Rlm ) (J p,lm J p,m J q,lm J q,m )(Rmm Rlm ) J p ,lm n J p ,i ( Rli Rmi ) q ,i ( Rli Rmi )]} 10 3 , i l i l ,m (1) J q ,lm n J (5) i l i l ,m (2) where t is the transpose symbol; R is the bus resistance matrix, which can be obtained from the corresponding bus impedance matrix Y by its inversion [7]. Let us suppose that we have load redistribution where Rll and R mm are proper resistances of the buses l and m; Rlm , Rli , and Rmi are mutual resistances of the buses l and m, l and i, m and i, respectively. This approach needs separate elements of the matrices R , which can be obtained without inversion of the matrices Y , using distribution coefficients and influence resistances [8]. The load homogeneity that may take place in distribution systems permits one to simplify (5): 2 (Plm ) 3{J lm ( Rll Rmm 2Rlm ) 2J lm [ J l ( Rll Rlm ) J m ( Rmm Rlm ) n J (R i li Rmi )]} 10 3 . (6) i 1 i l ,m Besides, the lack of reliable information in some cases allows one to be oriented only to using (6). Thus, the expressions (5) and (6) can serve for constructing exact equivalents to estimate the power system reaction to transferring the location of disconnection of a distribution network loop. Our experience shows that the use of experimental design [9] allows one to build functionally oriented models (their structure may be prompted, for example, by the structure of (5) or (6)) in rational way: the goal of experimental design is to organize experiments with a system or its model so as to maximize the amount of information obtained from a minimal number of experiments while simultaneously allowing a statistical evaluation of the result reliability. This evaluation provides: - the possibility to eliminate from consideration such factors, which have no essential influence on ( Plm ) to reduce the dimension of equivalents; - the possibility to test the adequacy of built equivalents and, if necessary, to change intervals of parameter varying to obtain adequate models. The experimental design technique is based on varying factors on a limited number of levels. In particular, a full factorial experiment is associated with conducting experiments for all combinations of factor levels. It is common to use the full factorial experiment with varying factors on two levels, that demands the performance of N 2 s experiments to construct models of the following type: y b0 s bp x p p 1 the replacements (g) defines the 2 s g experimental design. For example, to build the model ~ ~ ~ ~ (8) y b b ~ x b ~ x b ~ x , 0 1 1 2 2 3 3 we have to perform eight experiments, although it is enough to perform four experiments in accordance with the 2 2 full factorial design taking x 3 x1 x 2 if this interaction is insignificant. To illustrate the construction of exact equivalents as well as approximate equivalents (based on the use of experimental design) we consider the 138 kV power system shown in Fig. 1. Line and transformer (13813.8 kV) data are listed in Table 1. The load curves of feed system buses are given in Table 2 that also includes information about power ratings of the corresponding transformers. Let us consider the construction of (P2,6 ) . The elements of the matrix R necessary to construct (P2,6 ) on the basis of (5) or (6) are the R2,1 1.85 , R2,2 = 5.83 , R2,3 1.33 , R2, 4 3.82 , R2,5 3.82 , R2,6 R6,2 1.33 , R6,1 0.99 , R6,3 1.96 , R6,4 1.33 , R6,5 1.33 , and R6,6 8.74 . Then, on the basis of (6), for example, we can obtain following: (P2,6 ) (35.74 J 22,6 27.00 J 2,6 44.46 J 2,6 J 6 5.16J 2,6 J1 3.78J 2,6 J 3 14.94J 2,6 J 4 14.94 J 2,6 J 5 ) 10 3 . 1 2 1´ 2´ (9) 5 4 4´ 5´ s b pq x p x q 0 pq s b interaction effects of little significance (for example, x p x q x r in (7)) by new parameters. The number of pqr x p x q x r ... . (7) 3´ pqr Considering that 2 s 1, data obtained in the full experiment have excessiveness, which allows one to construct models on the basis of so-called fractional experiment. Its matrices are parts of the corresponding full experiment matrices, and a number of experiments, which must be performed, is less than a number of points in a factorial space. The fractional experiment matrices [9] may be obtained with reducing a number of experiments of the full experiment in two, four, etc. times, replacing 6´ s 3 6 Fig 1. Power system While constructing (P2,6 ) on the basis of experimental design, the following factors, defined by the structure of (6), are taken into account: J 2,6 , J 2,6 J 2 , J 2,6 J 6 , J 2,6 J 1 , J 2,6 J 3 , J 2,6 J 4 , and J 2,6 J 5 . Considering this, it is possible to apply the 2 74 fractional design [9] x1 J 2,6 , J 2,6 J 1 , x1 x3 taking x2 J 2,6 J 2 , x3 J 2,6 J 6 , x1 x 2 J 2,6 J 3 , x 2 x 3 J 2,6 J 4 , and x1x2 x3 J 2,6 J5. When analyzing models of multicriteria optimization, a vector F ( X ) = {F1 ( X ),..., Fq ( X )} of objective Table 1. Line and transformer data Line/Transformer 0 - 1´ 0 - 3´ 1´- 2´ 1´- 3´ 2´- 3´ 2´- 4´ 4´- 5´ 3´- 6´ 1´- 1 2´- 2 3´- 3 4´- 4 5´ - 5 6´ - 6 R ( ) 3.24 2.65 3.34 5.73 6.11 3.63 3.82 4.77 2.01 2.01 13.74 4.71 4.71 2.01 X ( ) 14.97 6.89 8.61 14.76 15.74 9.35 9.84 12.30 67.15 67.15 228.37 188.86 188.86 67.15 functions is considered, and the problem consists in simultaneous optimizing all objective functions, i. e., Fp ( X ) min , p 1,..., q , X L Table 2. Load curves of feed buses Transformer Bus Power Rating (MVA) 0-4 4-8 1 41 50 60 2 41 50 60 3 10 20 12 4 15 20 20 5 15 20 35 6 41 40 45 Load Curve (A) 8-12 12-16 16-20 20-24 110 100 120 70 100 90 100 50 15 20 30 30 30 55 50 25 50 30 50 40 90 90 80 50 To cover daily ranges of feed bus load changes with the sufficient margin (20%), the intervals: 40.0 J 1 144.0, 40.0 J 2 120.0, 9.6 J 3 36.0, 16.0 J 4 66.0, 16.0 J 5 60.0, and 32.0 J 6 108.0 have been determined. Besides, it has been taken 0.1 J 2,6 2.1. These load levels have served for constructing N 8 (in accordance with the 2 74 fractional design) combinations of 0.10 x1 2.10, 4.00 x 2 252.00, 3.20 x 3 226.80, 4.00 x1 x 2 302.40, 0.96 x1 x3 75.60, 1.60 x 2 x 3 138.60, and 1.60 x1 x 2 x 3 126.00 to realize the corresponding calculations of power system losses. As a result, the following model has been obtained: (P2,6 ) (6.87 78.60 J 2,6 26.98J 2,6 J 2 44.44J 2,6 J 6 5.16J 2,6 J1 3.78J 2,6 J 3 14.94 J 2,6 J 4 14.94 J 2,6 J 5 ) 10 3 . 3 MULTICRITERIA OPTIMIZATION OF NETWORK CONFIGURATION (10) The use of the models (9) and (10) leads to practically identical results. However, the way of constructing the power system reaction equivalents should be defined at every concrete case. (11) where L is a feasible region in R n . The lack of clarity in the concept of "optimal solution" is the fundamental difficulty in solving multicriteria problems. When applying the BellmanZadeh approach [10], this concept is defined with reasonable validity: the maximum degree of implementing all goals serves as a criterion of optimality. This conforms to the principle of guaranteed result and provides a constructive line in obtaining harmonious solutions from the Pareto set [11,12]. Besides, the approach permits one to realize a computationally effective method of analyzing models [11]. Finally, the approach allows one to preserve a natural measure of uncertainly in the process of decision making and consider indices, criteria, and constraints of qualitative (semantic, contextual) character based on experience, knowledge, and intuition of a decision maker (DM). Taking this into account, the approach has found wide applications in power industry problems [13]. When using the Bellman-Zadeh approach, each objective function F p ( X ), X L, p 1,..., q is to be replaced by a fuzzy objective function or a fuzzy set A p {X , Ap ( X )}, X L, p 1,..., q , (12) where Ap (X ) is a membership function of Ap [10]. A fuzzy solution D with setting up the fuzzy sets (12) is turned out as a result of the intersection q D A p with a membership function p 1 D ( X ) min A p ( X ), p 1,.., q X L . (13) With the use of (13) it is possible to obtain the solution X 0 providing the maximum degree of belonging to the fuzzy solution D max D ( X ) max min A p ( X ) (14) X L p 1,.., q D and reduced the problem (11) to X 0 arg max min A p ( X ) . X L p 1,..., q (15) To obtain (15), it is necessary to construct membership functions Ap (X ), p 1,..., q reflecting a degree of achieving "own" optima by F p (X ), X L , p 1,..., q . This condition is satisfied by the use of membership functions max F ( X ) F ( X ) p p p Ap ( X ) . (16) max F p ( X ) min F p ( X ) X L X L In specific cases it is possible to use min F p ( X ) A p ( X ) X L F p ( X ) p . (17) The peculiarities of solving the problem (14) with the use of the algorithms indicated above consists in the following. If X (m) is a current point, the transition to a point X ( m1) is expedient if (p 1,..., q) : Ap ( X ( m1) ) min Ap ( X ( m) ) . (21) 1 p q In contrast, if (p 1,..., q) : Ap ( X ( m1) ) min Ap ( X ( m) ) , (22) 1 p q ( m 1) the transition to X is not expedient. This way of evaluating the expediency of the transition to the next point X ( m1) leads to the solution (15) that is Pareto, if all inexpedient transitions are rejected. In (16) and (17) p , p 1,..., q are objective function importance factors. Their forming and correcting on the basis of a procedure convenient for the DM is considered in [13]. The construction of (16) demands to solve the following problems: (18) Fp ( X ) min , X L F p ( X ) max , (19) X L providing X 0p arg min F p ( X ) X L and X 00 p arg max F p ( X ), respectively. XL At the same time, the construction of (17) demands to solve only the problem (18). Hence, the solution of the problem (11) demands analysis of 2q+1 monocritéria problems (18), (19), and (14), respectively, or q+1 monocriteria problems (18) and (14), respectively. Since the solution X 0 is to belong to the Pareto set L , it is necessary to build D ( X ) min { min A ( X ), ( X )} , p 1,.., q p (20) where ( X ) 1 if X and ( X ) 0 if X . When analyzing problems of optimizing network configuration in multicriteria statement (as objective functions may be considered power losses, energy losses, undersupply energy, poor energy quality consumption, integrated overload of network elements, etc. in diverse combinations), several modifications of the univariate method [14] are used to solve the problems (18), (19), and (14). The corresponding algorithms (simple search, search for an effective coordinate, search for an effective step, etc.) are flexible and easily adapted to different practical strategies in the problem solution. 4 CONSIDERATION OF UNCERTAINTY FACTOR The consideration of the uncertainty factor is associated with constructing sets of rational solutions (solutions which belong to decision uncertainty regions or solutions which cannot be distinguished from the point of view of the considered objective function or considered objective functions) and their subsequent analysis to choose a final solution. Considering available initial information (loads, load curves, reliability indices, etc.), it is possible to form some representative combinations of initial data (so called "nature states") applying techniques considered in [15] or LPτ consequence techniques [16]. Resolving the corresponding monocriteria or multicriteria problem for each "nature state", it is possible to form the decision uncertainty region. The first step in analyzing the decision uncertainty regions is associated with applying the approach [15] based on elements of game theory. This approach consists in constructing and analyzing so called "payment matrices", which reflect effects obtained for different solution alternatives in accordance with formed representative combinations of initial data. The analysis of the decision uncertainty regions is based on applying special criteria [15]. The second step is associated with constructing and analyzing so called X , R models, where R {R1 ,..., Rt } is the vector fuzzy preference relation [11,17]. In this case, we have Rs [ X X , Rs ( X k , X l )] , s 1,..., t , X k , X l X , (23) where Rs ( X k , X l ) is a membership function of fuzzy preference relation. In (23), R s (also called a nonstrict fuzzy preference relation, fuzzy week preference relation, and fuzzy binary relation in literature) is defined as a fuzzy set of all pairs of the Cartesian product X X , such that the membership function Rs ( X k , X l ) represents the degree to which X k weakly dominates X l , i.e., the degree to which X k is at least as good as X l ( X k is not worse than X l ) for the sth criterion. In a somewhat loose sense [18], Rs ( X k , X l ) also represents the degree of truth of the statement " X k is preferred over X l ". Information given in the form (23) permits one to realize the evaluation, comparison, choice, and/or ordering of alternatives on the basis of criteria of quantitative as well as qualitative ("flexibility of operation", "comfort of maintenance", etc.) character. The questions of constructing and analyzing X , R models are considered in [11,19]. [6] [7] [8] [9] [10] [11] 5 Conclusion The results of research related to improving the validity and efficiency in solving problems of network reconfiguration in distribution systems have been presented. Techniques for building power system functional equivalents have been considered and illustrated. Taking into account multicriteria nature of the problems, they have been formulated and solved within the framework of multicriteria optimization models. Two approaches based on elements of game theory and fuzzy preference relations have been described to consider the uncertainty of information in solving the problems. [12] References: [1] M. Baran and F. Wu, Network Reconfiguration in Distribution Systems for Loss Reduction and Load Balancing, IEEE Trans. Power Delivery, Vol.4, No.2, 1989, pp. 1401-1407. [2] T. Taylor and D. Lubkeman, Implementation of Heuristic Search Strategy for Distribution Feeder Reconfiguration, IEEE Trans. Power Delivery, Vol.5, No.1, 1990, pp. 239-246. [3] E. López, H. Opazo, L. García, and P. Bastard, Online Reconfiguration Considering Variability Demand: Application to Real Networks, IEEE Trans. Power Systems, Vol.19, No.1, 2004, pp. 549-533. [4] M. Arias-Albornoz and H. Sanhueza-Hardy, Distribution Network Configuration for Minimum Energy Supply Cost, IEEE Trans. Power Systems, Vol.19, No.1, 2004, pp. 538542. [5] P. Ekel and V. Popov, Consideration of the [15] [13] [14] [16] [17] [18] [19] Uncertainty Factor in Problems of Modelling and Optimizing Electrical Networks, Power Engineering, Vol.23, No.2, 1985, pp. 45-52. P.Ya. Ekel, Methods of Decision Making in Fuzzy Environment and Their Applications, Nonlinear Analysis, Vol.47, No.5, 2001, pp. 979-990. O Elgerd, Electric Energy Systems. McGrow, 1975. V.G. Kholmsky, Calculation and Optimization of Electrical Network Modes of Operation, Visshaya Shkola, 1975 (in Russian). R. Jain, The Art of Computer System Performance Analysis: Techniques for Experimental Design, Measurement, Simulation, and Modeling, Wiley, 1991. W. Pedrycz and F. Gomide, An Introduction to Fuzzy Sets: Analysis and Design, MIT, 1998. P.Ya. Ekel, Fuzzy Sets and Models of Decision Making, Computers and Mathematics with Applications, Vol.44, No.7, 2002, pp. 863-875. P.Ya. Ekel and E.A. Galperin, Box-triangular multiobjective linear programs for resource allocation with application to load management and energy market problems, Mathematical and Computer Modelling, Vol.37, No.1, 2003, pp. 1-17. R.C. Berredo, R.L.J. Carnevalli, P.Ya. Ekel, and J.G. Pereira Jr., Fuzzy set based multiobjective allocation of resources and its power engineering applications, Proc. XXXVI Brazilian Symposium on Operations Research, São João Del-Rei, 2004, 1066-1075. S. Rao, Engineering Optimization: Theory and Practice, Wiley, 1996. L.S. Belyaev, A Practical Approach to Choosing Alternative Solutions to Complex Optimization Problems under Uncertainty, IIASA, 1977. I.M. Sobol', On the Systematic Search in a Hypercube, SIAM Journal on Numerical Analysis, Vol.16, No.5, 1979, pp. 790-793. P. Ekel, W. Pedrycz, and R. Schinzinger, A General Approach to Solving a Wide Class of Fuzzy Optimization Problems, Fuzzy Sets and Systems, Vol.97, No.1, 1998, pp. 49-66. P. Kulshreshtha, and B. Shekar, Interrelations Among Fuzzy Preference-Based Choice Functions and Significance of Rationality Conditions, Fuzzy Sets and Systems, Vol.109, No.3, 2000, pp. 429-445. R.C. Berredo, P.Ya. Ekel, and R.M. Palhares, Fuzzy Preference Relations in Models of Decision Making, Nonlinear Analysis, accepted.