Properly Learned Convolution Neural Networks and Local Optimization for Optimal Power Flow

Abstract

Recently, costs of power generation of power plants have risen greatly since energy has been lacked of and environmental protection greatly emphasized. The costs could not be transferred to the clients. Therefore, it is very important to apply optimal power flow to analysis of power market economy. In the paper, local optimization algorithm and convolution neural network are applied to determine solution of optimal power flow containing with both discrete and control variables based on Equivalent Current-Injection method. Based on the load flow model of the current injection method, the programmed variables are divided into continuous control variables and discrete ones. The continuous variables are generator-bus voltage magnitudes on the bus bar and the active power outputs in the ordinary optimized power flow method. The discrete variables are the shunt capacitor devices, transformer-tap settings and phase shifters. Next, convolution neural network properly learned to fast compute planning results. They are used as initial conditions for local optimization method. The proposed Artificial Intelligent method is utilized to verify for modify IEEE 30 bus power system and compared with those from other algorithms for faster computation efficiency and higher quality solution.