Short-term power load forecasting based on ARIMA-LSTM

Abstract

Accurate short-term forecasting of power load demand has become increasingly crucial in the field of electric systems due to the continuous development of society. This paper utilizes the ARIMA and the LSTM algorithm to forecast the future 24-hour electricity load of a region. The coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and mean absolute precision (MAPE) are utilized as model performance indexes. The prediction results for this data show indicator values of 0.843, 576.233, 435.423, and 6.964 for the ARIMA model, and 0.970, 243.86, 185.80, and 2.859 for the LSTM method. An improved ARIMA-LSTM model is then proposed, with weights assigned based on the magnitude of the errors predicted by the two models in the training set. Finally, the prediction results are obtained using the weights. The index values of the improved method are 0.9732, 241.39, 36.78, and 2.828. These indicate that the ARIMA-LSTM model outperforms the first two models.