Studies On Distributed Multi-Objective Reactive Power Optimization In Power Systems

With the expansion of power networks and the development of UHV AC/DC power grid and distributed generation technology, various calculations of power systems are becoming more and more complex. Thus, many traditional calculation methodes need to be improved for the development situation. As an important calculation for the safety and economic operation of power systems, reactive power optimization is a mixed nonlinear optimization problem with a large number of variables and constrains. Because of the above results, it is becoming more and more difficult to get a satisfying global solution for a large-scale power system using centralized optimization approaches. Considering reactive power needs to be compensated locally, a large-scale power system can be divided into several small-scale subsystems in network partition. Therefore the whole optimization problem is decomposed into several low-dimension and low-coupling subproblems for corresponding subsystems. Then the optimization problem is solved coordinatively in distributed computing. It is of great significance that the computing mode conforms to the development of power systems.Introducing the theory and methods of fuzzy clustering, fuzzy evaluation, multi-agent system and intelligent optimization, distributed multi-objective reacitive power optimization is researched systematically in the dissertation. The main contributions of the dissertation are shown as follows.A new network partition approach based on the improved fuzzy C means clustering algorithm is presented herein. In order to take into account the perturbation impact of discrete variables on nodes' parameters, such as transformer tap-changers and capacitor banks, a perturbation approach is adopted to get the voltage response of nodes to the perturbation of various discrete and continuous variables. The approach takes into account the combined effect of reactive power variables in perturbation analysis, which overcomes disadvantages of traditional sensitivity approaches. The nodes in power networks are mapped to the space of var variables according to the normalized perturbation values, and a new electrical distance is worked out associated with an improved incidence matrix. Therefore the electrical distance combines network topology information with the voltage response of nodes, which can avoid the unreasonable partition results. And then an improved fuzzy C means clustering algorithm is used for network partition, in which the clustering parameters are initialized according to a improved a decomposition algorithm or dispatching areas. The effective initial parameters can avoid trapping into local optimal solution of clustering problems. In the clustering process, a clustering validity index is defined to evaluate the clustering results and confirm the final result of network partition. And expert knowledge is also introduced into the clustering algorithm for the reasonableness of partition results. The fuzzy clustering algorithm is a "flexible" method for nerwork partition, which can reflect more objectively the uncertainty of the classification and increase the probability of getting the global optimal solution of network partition. The simulations verify that the reasonable and effective results can be obtained using the proposed partition method.By the above method of network partition, a reactive power optimization problem can be decomposed into several subproblems of small-scale systems, which can reduce the complexity of var optimization problem significantly. A distributed multi-objective reactive power optimization method based on the multi-agent technology is presented in the dissertation. Considering the characteristics of reactive power optimization in different network structures, two kinds of agents are developed according to network structures: loop agent for loop network and radial agent for radial network. The characteristics of two kinds of agents for reactive power optimization are presented respectively in power flow algorithm, objective function and optimization algorithm. Membership functions for the distributed multi-objective optimization problem are constructed to evaluate objectives, so that the objectives with different units can be compared. Evaluation values of multi-objective optimization problem are regarded as coordinate values of points in a multidimensional space, and the Euclidean distances between the points and the ideal point are used to evaluate optimization solutions. According to the decomposition and coordination theory, agents transmit the parameters of boundary nodes each other via Internet in the process of distributed var optimization. Due to the unobservability of internal data among subsystems, the integrity and sealing of data is ensured. In addition, a grid-computing architecture for distributed multi-objective reactive power optimization is designed based on open grid service architecture. Grid computing technology can solve the data-sharing problem of distributed resources and integrate the heterogeneous computational resources. The optimization subproblems for corresponding subsystems can be worked out autonomously by agents, which are wrapped into grid services. The simulation results, greatly reduced power losses and improved voltage profiles as well as coordinated the parameters of boundary nodes, show that the method is feasible and effective.The solutions of reactive power optimization not only meet the requirement of power system economic operation, but also meet the requirement of voltage security. The improper allocation of reactive power resources will have hidden dangers of voltage instability accidents in power systems, thus it is necessary to research the reactive power optimization considering voltage stability. For considering voltage stability of power transmission paths in power systems, a new distributed reactive power optimization considering voltage stability method is introduced in the dissertation. The local voltage stability index as the optimization objective is used to improve the stability margin of weak nodes after fuzzy evaluation. Then according to optimization results and weak nodes, the voltage stability indeces of weak paths are computed and the weakest power transmission path is searched. The reactive power reserve index of key power sources is developed from the key generators reactive power reserve index, including virtual power sources as well as key generators. The two kinds of indeces are incorporated as coordinate values in a four-zone diagram, which is used to evaluate the solutions of reactive power optimization. Weak nodes and weak paths of voltage stability in subsystems are considered in the proposed method of reactive power optimization. The results of IEEE 30-bus system in the base and heavy load conditions show that the method can ensure the voltage stability margin of power systems and avoid accidents of voltage instability due to the improper allocation of reactive power resoures.