Combined model electricity load forecasting based on NeuralProphet and Bi-LSTM-SA

Abstract

Abstract This study proposes an innovative method for forecasting electricity load that combines NeuralProphet’s time series analysis capability with Bi-LSTM-SA’s self-attention mechanism. The method improves prediction accuracy, reliability, and interpretability by analyzing trends, cycles, and holiday impacts, as well as considering climatic factors as key external variables. A peak interval weighted mean square error indicator is introduced to optimize the weights in the model combination strategy. This improves the prediction accuracy during peak times, making this method superior to any single sub-model in terms of prediction performance.