Publication

Abstract :

Accurate short-term load forecasting (STLF) is crucial for the planning and operation modern power systems, particularly in regard to increasing energy demand variability and the integration of renewable sources. This work presents a hybrid deep learning architecture which utilises the Fourier Enhanced Decomposed Transformer (FEDFormer) with the Feature-Enhanced Cross Attention Module (FECAM) to address the problem of temporal complexity and feature relevance in power consumption data. FEDFormer identifies global patterns by frequency-aware decomposition, while FECAM highlight critical elements through cross-attention techniques. Leveraging the UCI Household Electric Power Consumption dataset, our model achieves an RMSE of 0.2243 and an MAE of 0.101 showing reduced error compared to conventional models and existing deep learning equivalents. The results confirm that the proposed hybrid model provides a scalable and robust framework for practical short-term load forecasting applications.