Multi-Objective Distribution Network Reconfiguration Based on Deep Learning Algorithm

摘要

Distribution network reconfiguration is an important means to improve power supply reliability and reduce network loss. In this paper, a deep learning CNN (convolution neural network) model is established to solve the problem of switch status optimization under the constraints of distribution network operation. Deep learning is a branch of machine learning, for some complex issues, deep learning CNN can autonomously combine basic features into more complex features and learning the rules to solve practical problems. Firstly, a distribution network reconfiguration model with multi-objective as active power loss, average power supply availability indicator and load balancing indicator of the system is established. With the judgement matrix, the weight of each target is optimized according to the expert experience, and the multi-objective is transformed into a single-objective. Then, under different loads situation, the optimization model of single-objective distribution network reconfiguration model is solved by the particle swarm optimization algorithm to get the optimized switch open/closed combination. Taking the load of each node as input, the optimized switch open/closed combination is output, trained in deep learning CNN model, and the load characteristic is extracted by deep convolution neural network to simulate the nonlinear relationship of reconfiguration. The trained deep learning model can well simulate the switch status combinations under different loads and meet the requests of the optimization objective without Iteration and improve the reconfiguration efficiency in the actual distribution network.