Neuro-Wavelet Approach to Time-Series Signals Prediction: An Example of Electricity Load and Pool-Price Data

Accurate electricity load and pool-price forecasting can provide a set of vital predicted information that helps generation, transmission and retailer participating companies to bid strategically into a deregulated electricity market in order to maximize their profits and increase returns to their stakeholders. Although a number of forecasting methods have been proposed to solve the short-term and long-term electricity load forecast, pool-price forecasting is a relatively new research area. In this article, we propose an autoregressive approach, based on a wavelet multiscale decomposition, for the prediction of one-hour ahead load and pool price based respectively on historical electricity load, and pool-price data. This approach is based on a multiple resolution decomposition of the signal using the redundant Haar à trous wavelet transform whose advantage is taking into account the asymmetric nature of the time-varying data. There is an additional computational advantage in that there is no need to re-compute the wavelet transform (wavelet coefficients) of the full signal if the electricity and pool price data (time series) is regularly updated. We assess results produced by this multiscale autoregressive method, in both linear and nonlinear variants, with single resolution autoregressive, multilayer perceptron, Elman recurrent neural network and the general regression neural network models. The input data consists of historical load and pool price data, which is collected over a period of 3 years (1999-2001), used for training, and 1 year (2002) used for testing. Experimental results are based on the New South Wales (Australia) electricity load and pool price data that is provided by the National Electricity Market Management Company.