Cascaded Ensemble-based Short-term Load Forecasting for Smart Energy Management

The rising adoption of renewable energy generation coupled with the anticipated increase in demand for reliable electricity over the coming years has brought attention to the importance of accurate short-term load forecasting. Short-term load forecasting plays an essential role in the scheduling and planning of the power grid's resources to ensure it operates efficiently and reliably. This paper proposes a short-term load prediction model that exploits XGboost and LightGBM models under a cascaded ensemble architecture to provide highly accurate predictions. The architecture minimizes the weaknesses of individual predictors by combining advanced feature engineering and feature selection strategies. The proposed model's effectiveness is tested on the publicly available benchmark dataset using mean average percent error (MAPE) to evaluate the models accuracy, while runtime is used to evaluate the model's computational efficiency. The proposed model demonstrates increased performance when compared to both the baseline model and conventional models.