Short-term load forecasting based on improved optimization algorithms

Accurate predictions of electricity load are crucial for the safe and stable operation, as well as the economic performance, of power systems. To overcome the inadequacy of feature selection and the difficulty in tuning model hyperparameters in traditional load forecasting models — issues that may lead the predictive model into local optima — this paper proposes a short-term electricity load forecasting method based on XGBoost-IARO-LSTM.Initially, the eXtreme Gradient Boosting (XGBoost) tree model is employed for feature selection. Subsequently, the Improved Artificial Rabbits Optimization (IARO) algorithm is enhanced using tent chaotic mapping, elite reverse learning strategy, and hybrid reverse learning strategy. The improved IARO algorithm is then utilized to tune the hyperparameters of Long Short-Term Memory (LSTM).Finally, the XGBoost-IARO-LSTM combined forecasting model is established to achieve accurate short-term load predictions. The proposed method is validated using real electricity load data from a specific region in northern China. The results indicate that the proposed approach exhibits faster parameter optimization speed and superior prediction performance.