Research and optimization of large-scale power load forecasting model based on transformer

With the continuous expansion of power system scale, power load sequences exhibit significant nonlinearity, strong periodicity, and multi-scale fluctuation characteristics. Traditional forecasting methods are gradually failing to meet the requirements of large-scale applications in terms of accuracy and stability. To address the long-standing problem of limited modeling and computational efficiency in complex load scenarios, a large-scale power load forecasting method based on the Transformer architecture is constructed. Results show that this method achieves a MAE of 3.86 MW, an RMSE of 5.94 MW, and a MAPE reduced to 4.37% on the test set, with a determination coefficient reaching 0.958, demonstrating significant improvements over various comparative models. In cross-regional generalization tests, the NRMSE decreases to 0.098, the prediction interval coverage increases to 0.94, and error fluctuations converge significantly, maintaining stable forecasting performance in both peak and off-peak load intervals. This method demonstrates good advantages in forecasting accuracy, generalization ability, and stability, providing a practical technical solution for large-scale power load forecasting.