Enhancing Short-term Electric Load Forecasting with EMD-GA Optimized LSTM

To enhance the accuracy of load forecasting, this paper considers both the inherent patterns within historical load data and external influencing factors. A short-term load forecasting approach is proposed that combines Empirical Mode Decomposition (EMD), Genetic Algorithm (GA), and Long Short-Term Memory (LSTM). Inputs include weather conditions, temperature, hourly weights, and date weights, while load data is the output. Historical data from a building in an industrial park in the Wuhan region is utilized as the dataset. Various prediction models are compared, including Back Propagation Neural Network (BP), Support Vector Regression (SVR), Random Forest (RF), and Long Short-Term Memory (LSTM). The evaluation metrics (MAE, RMSE, R
²
) indicate that LSTM outperforms the others. To further enhance the LSTM model, Empirical Mode Decomposition (EMD) is introduced for data preprocessing, and Genetic Algorithm (GA) is employed to optimize LSTM’s hyperparameters. The results show that the EMD-GA-LSTM model achieves the best performance in load forecasting with MAE, RMSE, and R
²
values of 6.18, 7.31, and 0.994, respectively, which are better than the other compared methods.