Bayesian neural network model to predict day‐ahead electricity prices

Abstract

AbstractMarket participants in the new competitive electricity markets find price forecasting so valuable in developing their bidding strategies. In this paper, the Bayesian neural network (BNN) approach is proposed for day‐ahead prediction of locational marginal prices (LMPs) in the pool based energy markets, in which to select the optimum inputs of the BNN, the correlation coefficient technique is used. LMP has a volatile and time dependent behavior. Hence, prediction of such a complex signal is a challenging task requiring a qualified forecasting tool, which not only fits well to the training data, but also can predict the stochastic behavior of unseen part of the signal. The proposed Bayesian approach to predict day‐ahead electricity prices presents noteworthy advantages over the classical neural networks (NN) methods which include the avoidance of network overfitting, an indication of the degree of uncertainty in the predictions, automatic selection of an appropriate scale for network weights and, accordingly, selection of the optimal forecasting model. Examining the proposed forecast strategy on the different periods of the Pennsylvania‐New Jersey‐Maryland (PJM) market, it is demonstrated that the proposed method can provide more accurate results than the other price forecasting techniques, such as, ARIMA time series, wavelet‐ARIMA, classical NN, and also a similar day method. Copyright © 2008 John Wiley & Sons, Ltd.