Ensemble models of TCN-LSTM-LightGBM based on ensemble learning methods for short-term electrical load forecasting

The accurate forecasting of electrical loads is essential for optimizing energy dispatch and reducing expenses. In this study, a novel ensemble model of a temporal convolutional network-long short-term memory-light gradient-boosting machine (TCN-LSTM-LightGBM) for short-term power-load forecasting is proposed. Multiple linear regression is used to integrate the outputs of the TCN-LSTM and LightGBM models. The predictive performance of the proposed model is evaluated using two datasets from Australia and China. In addition, the performance of the ensemble model is compared under different ensemble methods. The results show that, except for the dates with significant random load changes, the proposed ensemble model has good prediction capability compared to the other models. In the Australian loads dataset, the mean absolute error (MAE) of the ensemble TCN-LSTM-LightGBM model is reduced by an average of 4.88% and 28.95% compared to the TCN-LSTM and LightGBM models, respectively, under the four typical days. Compared to other ensemble methods, the multiple linear regression ensemble method proposed in this study reduces the MAE of the hybrid model by an average of 3.64% and the root mean square error by an average of 2.44%. The research results have significant reference value for improving the predictive performance of ensemble models.