Modeling and Forecasting Electricity Price Based on Multi Resolution Analysis and Dynamic Neural Networks

Background

The purpose of this study is to forecast California electricity price using artificial neural networks (ANN) and continuous wavelet transform (CWT) as multi resolution technique to analyze the original data to extract hidden patterns. In particular, the conventional backpropagation neural network (BPNN), the Elman recurrent neural network, the time delay recurrent neural network (TDNN), and the adaptive time delay recurrent neural network (ATDNN) were used as predictive systems. The TDNN and ATDNN integrate genetic algorithms (GA) to optimize their respective architecture.