Enhancing Electric Load Forecasting with Automated Machine Learning: An Approach for Improved Accuracy and Efficiency

Accurate electric load forecasting is crucial for the efficient management of power systems, ensuring reliability and optimal energy distribution. Traditional forecasting methods often require extensive domain expertise and manual intervention. This study investigates the application of Automated Machine Learning (AutoML) to enhance electric load forecasting. By leveraging AutoML, we optimize and refine model selection and hyperparameter tuning, to improve forecasting accuracy with minimal human effort. The default 24 models in AutoML (Model 1), using a weighted ensemble, achieve a MAPE of 11.68 % and RMSE of 8.41 kW but require the longest training time of 1653 seconds. The 'Select Best Models' (Model 2), also using a weighted ensemble, offers slightly lower accuracy (MAPE of 13.99 %, RMSE of 8.08 kW) with a faster training time of 1019 seconds. Fine-Tuning hyperparameters in the best models (Model 3), with a weighted ensemble, delivers the best overall performance (MAPE of 13.54 %, RMSE of 7.21 kW) with the shortest training time of 760 seconds, making it the optimal choice. Our experiments, conducted on real-world datasets, demonstrate that AutoML can significantly improve performance in both accuracy and computational efficiency.