Short-Term Electricity Load Forecasting Method Based on Attention and Time Series Forecasting

Abstract

Short-term electricity load forecasting is critical in power system operation and control. However, accurate prediction of electricity load is challenging due to its non-stationary and intermittent characteristics. This paper proposes a hybrid deep learning model optimized with the attention mechanism for short-term electricity load forecasting, named CNN-LSTM-Attention. CNN-LSTM-Attention uses temporal convolution and long short-term memory network to extract temporal features of electricity load data and uses attention mechanism to optimize the fusion of long-term temporal features. The empirical study uses real electricity consumption data from four seasons in two regions. The forecasting results of the proposed model outperform all benchmark methods, demonstrating the advanced forecasting ability of the CNN-LSTM-Attention model.