A new day-ahead hourly electricity price forecasting framework

Abstract

This paper develops a hybrid electricity price-forecasting framework to improve the accuracy of prediction. A novel clustering method is proposed that uses a modified game theoretic self-organizing map (GTSOM) and neural gas (NG) along with competitive Hebbian Learning (CHL) to provide a better vector quantization (VQ). To resolve the deficiency of the original SOM, five strategies are proposed to enable the non-winning neurons to participate in the learning phase. Using GTSOM, the price-load input data are clustered into proper number of subsets. A novel cluster-selection method is proposed to select the most appropriate subset whose time-series data is processed to provide the inputs for the neural networks. Finally, Bayesian method is used to train the networks and forecast the electricity price. Market price data from an independent system operator is used to evaluate the algorithm performance. Furthermore, a comparison of the proposed method against other state-of-the-art forecasting techniques shows a significant improvement in the accuracy of the price forecast.