Day-Ahead Marginal Price Forecasting Based on Autoregressive Conditional Heteroskedasticity-Back Propagation Network Model

An improved neural network model based on autoregressive conditional heteroskedasticity (ARCH) analysis is proposed for day-ahead electricity market. Firstly,by use of ARCH analysis the conditional variance of marginal price series is obtained; then taking the conditional variance as the risk index of price fluctuation,an ARCH-back propagation networks (BPN) model,which is based on historical prices,historical loads and conditional variances of historical prices,is built,and by use of the built model the day-ahead marginal prices of Pennsylvania-New Jersey-Maryland (PJM) electricity market in United States are forecasted. Forecasting results show that by means of leading in ARCH the forecasting accuracy of traditional BPN can be effectively improved.