Short-term Load Forecasting Based on Elman Neural Network Optimized by Bayesian Algorithm

The global energy structure is facing transformation, and the power system is constantly improving. Solving the problem of poor short-term load forecasting accuracy caused by artificially determining Elman neural network hyperparameters is one of the problems faced at present, the Bayesian Optimization Algorithm (BOA) is employed to identify the optimal combination of hyperparameters for the Elman neural network in this study. A novel BOA-Elman model for short-term load forecasting is then proposed, which allows for an analysis of the impact of hyperparameters on the model's prediction accuracy. BOA is employed to optimize the hyperparameters of the Elman network model in order to determine the best combination, this optimized model is then applied to conduct short-term load forecasting. The accuracy of the predictions is evaluated using specific evaluation metrics. To validate the effectiveness of the BOA-Elman prediction model, real load data from a specific location is utilized as a case study. A comparative analysis is conducted by comparing the BOA-Elman prediction model with the traditional Elman neural network model, LSTM neural network model, and BOA-LSTM model, the MAE is reduced by 56.35%, 44.89%, 40.29%, the RMSE is reduced by 57.75%, 49.15%, 43.28%, the MAPE decreased by 55.62%, 43.38%, 38.67%, and the R 2 increased by 21.47%, 15.56%, 7.87%, respectively. The fitting degree of BOA-Elman prediction model can reach 97.3%, and its prediction curve is more consistent with the real load data than the other three models. Compared to the other three models evaluated in this paper, the BOA-Elman prediction model demonstrates superior prediction accuracy and exhibits greater stability.