A Short-Term Power Load Forecasting Approach Based on EMD-SSA-LSTM

Abstract Accurate power load forecasting is the foundation of the power system. However, economic, weather, and geographic factors can affect short-term electric load forecasting and can make it difficult to enhance load forecasting accuracy. To solve the problem, a novel load forecasting method based on empirical mode decomposition (EMD) and long short-term memory (LSTM) neural networks has been proposed. Firstly, the EMD algorithm is utilized to decompose the original signal into elementary components and a series of intrinsic mode functions ( IMF ). Subsequently, all components are input into the LSTM neural networks for load prediction. Moreover, to further improve the prediction accuracy, the sparrow search algorithm (SSA) is introduced to optimize the parameters of LSTM for the prediction of basic components. Integrating SSA and EMD-LSTM, the EMD-SSA-LSTM model is further proposed for short-term power load forecasting. Finally, the proposed model is validated by several experiments, and the results show that the coefficient of determination (R 2 ) under the proposed model can reach 0.98, the mean absolute error (MAE) is 0.015, and the root mean square error (RMSE) is 0.02. It is verified that the proposed model can improve the accuracy of load forecasting.