Short-term power load forecasting based on sparrow search algorithm-variational mode decomposition and attention-long short-term memory

Abstract

Abstract To improve the forecasting accuracy of power load, the forecasting model based on sparrow search algorithm (SSA), variational mode decomposition (VMD), attention mechanism and long short-term memory (LSTM) was proposed. Firstly, SSA is used to optimize the number of decomposition and penalty factor in VMD and realize the decomposition operation of the initial data. Then, LSTM is used to predict each component, and on this basis, feature and temporal attention mechanisms are introduced. Feature attention mechanism is introduced to calculate the contribution rate of relevant input features in real time, and the feature weights are modified to avoid the limitations of traditional methods relying on the threshold of expert experience association rules. Temporal attention mechanism is applied to extract the historical key moments and improve the stability of the time series prediction effect. Finally, the final result is obtained by superimposing the prediction results of each component to complete the power load prediction. Practical examples show that, compared with other methods, the proposed model achieves the highest prediction accuracy, with an RMSE of 1.23, MAE of 0.99 and MAPE of 11.62%.