Publication

Abstract :

Effective short-term load forecasting is essential for maintaining grid stability and efficient resource allocation. As the meteorological and temporal features strongly influence energy generation, fluctuations in these parameters significantly affect the load patterns. To enhance the short-term load forecasting better, the proposed methodology utilizes the ERCOT load dataset and implemented the Gated Recurrent Unit (GRU) model. Leveraging its fewer gates, it reduces 19 % of total parameters which improves the computational efficiency compared to LSTM and still achieves an R2 score of 0.991. To enhance the interpretability of the model, Permutation SHAP is employed, providing insights into the importance of feature contributions to the predictions. Furthermore, this work also extends the forecasting horizon, GRU shows a competitive results and also benefits in parameter reduction, emphasizing it as a resourceefficient alternative model for short-term load forecasting applications and a robust decision-making support for grid users while experimenting with real-time data.