Decentralized voltage control in distribution system using neural network

Abstract

In recent years, distributed generation based on natural energy or using co-generation system is increasing due to the problems of global warming and exhaustion of fossil fuels. Many of the distributed generations are set up in the vicinity of the customer, with the advantage that this decreases transmission losses and transmission capacity. However, output power generated from natural energy such as wind power, photovoltaic generations, etc, which is distributed generation, is influenced by meteorological conditions. Therefore if the distributed generation increases with conventional control schemes, it is expected that the voltage variation of each node becomes a problem. This paper proposes a decentralized control of distribution voltage with distributed installations, such as load ratio control transformer (LRT), SSteptep voltage regulator (SVR), shunt capacitor (SC), shunt reactor (ShR), and static Var compensator (SVC). Neural network (NN) is used to determine the operation of the control device.The optimal data is created by genetic algorithm. By using the optimal data for training of NN, the operation of the control device can approach the optimal operation without the communication infrastructures. Furthermore, the decentralized control has the merit of robustness against faults of communication lines and local rapid voltage variation. In order to confirm the validity of the proposed method, simulations are carried out for a distribution network model with photovoltaic (PV) generators.