Weights Optimization Based on Genetic Algorithm for Variable Weight Combination Model of BP-LSSVM for Day-ahead Electricity Price Forecasting

Abstract

Affected by various factors, the electricity market prices are highly volatile seasonal and stochastic with nonlinear and non-stationary characteristics. The previous literature indicates that improving the accuracy of price forecasting by applying one model alone can be problematic .To evaluate the prices properly and efficiently, this study proposes genetic algorithm(GA) -back propagation neural network(BP)-least square support vector machine (LSSVM) (GA-BP-LSSVM) model. The original prices series are preprocessed to eliminate the outliers, then a BP-LSSVM is built to forecast the prices. In order to determine the weights of this combination model, GA is proposed to optimize the weights. Compared with traditional LSSVM, BPNN and BP-LSSVM, the forecasting results of this proposed model applied to the electricity market of Pennsylvania-New Jersey-Maryland (PJM) suggest the proposed model outperforms other models.