The application of improved BP neural network for power load forecasting in the island microgrid system

Abstract

Based on BP neural network, the paper set up an island microgrid system to forecast electricity load. In order to improve forecast accuracy and convergence speed, author updates BP neural network in two ways, on one hand, putting forward several limitations of the basic algorithm for BP, the paper firstly gives several commonly used improved BP algorithm to be compared and then select the LM algorithm as the subject of load forecasting network algorithm. On the other hand, the orthogonal least squares method is applied to select the variables on the accuracy impact for input, thereby seasonal classification is made on this historical data. A new method of data normalization and further data processing is proposed, it could therefore improve the design of neural network input layer. Compared to existed common network, the improved BP neural network could be able to serve better on fitting the original data, and owns a more substantial increase on the island load forecasting accuracy and convergence speed.