ELM-RBFN-based Deep Neural Network with Wavelet Transform of Level 4 and Ensemble Learning for Electricity Price Forecasting

This paper presents a new Deep Neural Network (DNN) method for forecasting electricity prices. The goal of this technique is to maximize profits and minimize risks in power markets. It employs an Autoencoder (AE)-based pre-learning approach of DNN in conjunction with an Extreme Learning Machine technique for Radial Basis Function Neural Network (ELM-RBFN). This method integrates ELM concepts into the RBFN framework. AE reduces the dimensionality of the input variables, while the ELM-RBFN enhances the performance of the RBFN through probabilistic techniques, effectively eliminating the need for clustering methods. Additionally, the paper utilizes a Level-4 Wavelet Transform as a feature extraction technique for the input variables. To further improve the performance of the proposed method, an ensemble technique is also employed. The method has been successfully applied to actual data from ISO New England in the USA.