Operating Pattern Based Distribution System Optimization

Distribution network optimization is the important part of distribution automation. The main methods of distribution network optimization are distribution network reconfiguration and capacitor switching. Both of them are the complicated optimization problem and the corresponding algorithms proposed have their advantages and disadvantages. Pattern analysis is introduced into distribution network optimization in this dissertation, which provides a new path to implement distribution network optimization.Distribution networks can be classified by homeomorphism graph depending on the property of distribution network, and each class is defined as a configuration pattern. Then the distance vector is defined to evaluate the difference of configuration patterns. A construction method of pattern recognition model for distribution reconfiguration problem is presented based on the configuration pattern, which can be used to solve the description difficulties of output space.Two distribution network reconfiguration algorithms are proposed based on the configuration pattern: one algorithm is based on improved optimal flow pattern algorithm and genetic algorithm (GA);the other is based on Support Vector Machine (SVM).The algorithm based on improved optimal flow pattern algorithm and genetic algorithm divides the global search space of the reconfiguration problem into several subspaces by use of configuration pattern. Then the improved optimal flow pattern is used to find out the local optimal solutions within the subspaces. At last the GA is applied to find out the subspace where the global optimal solution exists. Thereby, it is implemented to search global optimal solution among local optimal solutions. The algorithm can enhance the search efficiency of GA by compressing the search space and improve the local search ability by use of improved optimal flow pattern algorithm.The algorithm based on Support Vector Machine describes the distribution reconfiguration problem as a problem of pattern recognition, and presents a construction method of pattern recognition model based on configuration pattern of distribution network and SVM. The method takes full advantage of Support Vector Machine's ability to solve the problem with small sample, nonlinear and highdimension. Hence the method can obtain better generalization ability. Configuration pattern of distribution network is used to solve the description difficulties of output space in the model. The method proposed is applicable for the reconfiguration problem of medium scale distribution system.A static reconfiguration algorithm and a dynamic reconfiguration algorithm for distribution system based on time interval are proposed. The algorithms are based on optimal flow pattern and the objective function is the minimum of energy loss during a period of time. A modified calculation method of optimal flow pattern is presented based on the load pattern of distribution system. Then the method is used to solve the static reconfiguration optimal.The constraint of operation numbers for switches must be considered in dynamic reconfiguration optimal. The difference of switch operations, which are divided into pattern operations and non-pattern operations, is first analyzed, and then some effective heuristic rules are proposed according to the analysis. A formula for energy loss reduction is also proposed to calculate and estimate the effectiveness of different switch operations. The optimal operating sequence of switches can be determined based on these heuristic rules and the formula.The dynamic reactive power optimization algorithm of distribution network is proposed. Dynamic reactive power optimization of distribution system is to optimize switching time and size of capacitors under the consideration of load variations and the constraint of capacitors maximal allowable operation times. The algorithm decouples the problem into a series of single capacitor dynamic optimization sub-problems. The solutions of main problem can be obtained by the iterative solving of these sub-problems. A sub-algorithm to solve the single capacitor dynamic optimization sub-problem is proposed, which can get the optimal solutions under variant constraints, and both of the switching time and size of capacitors can be modified in the iteration, therefore the main problem is able to achieve better solutions.