Electricity Price Forecasting and Classification Through Wavelet–Dynamic Weighted PSO–FFNN Approach

Abstract

In competitive electricity markets, accurate price forecasting is required to both power producers and consumers for planning their bidding strategies in order to maximize their own benefits. Price classification is an alternative approach to forecasting where the exact values of future prices are not mandatory. Presently, two efficient algorithms are proposed for both short term price forecasting (STPF) and classification (STPC) purposes. The algorithms include various methodologies like wavelet transform (WT), fuzzy adaptive particle swarm optimization (FA-PSO) and feed forward neural networks (FFNN). WT is utilized to convert the pathetic price series to an inviolable price series without losing the originality in the signal. Standard PSO (S-PSO) is implemented to tune the fixed architecture FFNN weights and biases. In the present nonlinear problem, linear variation of inertia weight does not resemble exact search process. Hence, dynamic inertia weight is accomplished by implementing the fuzzy systems in the S-PSO approach. The hybrid methodology is implemented on Spanish electricity markets for the year 2002. To validate, three types of price classes and historical price series that are utilized by many researches as input features, are considered. Various statistical indicators are evaluated to compare and validate the proposed approaches with the past approaches available in the literature survey.