Short-Term Load Forecasting Based on Wavelet Transform and Chaotic Bat Optimization Algorithm-Long Short-Term Memory Neural Network

Abstract

To achieve an accurate forecast of short-term load, a new method via wavelet transform (WT) and chaotic bat optimization algorithm (CBA) and long short-term memory neural network (LSTM) is proposed. Firstly, the actual load data is decomposed by WT, and multiple groups of modal function components with different characteristics are obtained. Then the hidden neurons, initial learning rate, and iteration times in the LSTM regression model are optimized based on the CBA. Then the modal function components of each group are predicted respectively. Finally, the predicted modal component functions are reconstructed to achieve power load prediction. The results show that the WT-CBA-LSTM achieves accurate load prediction, with an average absolute error of 29.68 MW, a root mean square error of 52.14 MW, and an average absolute percentage error of 0.59%. However, the evaluation indicators of the traditional LSTM method, the WT-LSTM method and the CBA-LSTM method are greater than those of the WT-CBA-LSTM method, so the WT-CBA-LSTM has high accuracy in the process of load prediction.