A new short-term energy price forecasting method based on wavelet neural network

Abstract

A wavelet neural network (WNN) is proposed for short-term price forecasting (STPF) in electricity markets. Back propagation algorithm is used for training the wavelet neural network for prediction. Weights in the back propagation algorithm are usually initialised with small random values. If the random initial weights happen to be far from a suitable solution or near a poor local optimum, training may take a long time or get trapped in the local optimum. In this paper, we show that WNN has acceptable prediction properties compared to other forecasting techniques. We investigated proper weight initialisations of WNN, and proved that it attains a superior prediction performance. Finally, we used a two-step correlation analysis algorithm for input selecting. This algorithm selects the best relevant and non-redundant input features for WNN. Our model is examined for MCP prediction of the Spanish market and LMP forecasting in PJM (Pennsylvania, New Jersey and Maryland) market for the year 2002 and 2006 respectively.