Electricity price and load demand forecasting using an adaptive hybrid BiLSTM model based on wavelet transform technique and Pareto optimization: an application in the smart cities

Abstract

Accurate forecasting of electricity price and load demand is an essential requirement for managing energy production and consumption in a smart city. In this paper, an adaptive hybrid model is presented for accurate short-term forecasting of electricity price and load demand based on wavelet transform (WT) decomposition, mutual information (MI) and interaction gain (IG) feature selection methods, and Pareto optimization technique with BiLSTM network called WT-MI-IG-BiLSTM. In this model, first, the electricity price and load demand signals are decomposed using the WT technique. Then, the variables that have the most excellent effect on the prediction are selected by the MI and IG feature selection method. In the forecasting stage, prediction is made with the BiLSTM network, and the combination of networks prediction vectors provides the final prediction result. PJM electricity market price and load demand data sets in 2006 and 2018 and five error metrics including RMSE, MAE, MAPE, Variance, and R-Squared are used to evaluate the model. To demonstrate the high capability of the WT-MI-IG-BiLSTM model, the proposed model has been compared with the MI-IG-BiLSTM, WT-MI-IG-LSTM, and MI-IG-LSTM models. Based on the obtained results, the proposed WT-MI-IG-BiLSTM model compared to the MI-IG-BiLSTM model, which is the best benchmark model, has 17-18.16% improvement in accuracy of electricity price forecasting and 21.8% in accuracy of electricity load forecasting. Finally, the Pareto optimization algorithm has implemented on the model, and a set of optimal models with optimal accuracy and execution time has presented in the Pareto front chart.