ADEPT: Adaptive Energy Forecasting with Parameter-Efficient Tuning for Smart Grid Operations

Energy load forecasting faces significant challenges from concept drift caused by evolving weather patterns, consumer behaviors, and grid dynamics. We propose ADEPT (Adaptive Energy forecasting with Parameter-Efficient Tuning), a novel framework that addresses these challenges through energyspecific concept encoding, lightweight parameter adaptation, and asymmetric loss functions tailored for grid operations. ADEPT achieves $\mathbf{4. 2} \boldsymbol{\%}$ RMSE improvement over state-of-the-art methods on ECL dataset, reduces peak load forecasting errors by 38 %, and provides 28 % computational savings with 37 % faster adaptation times. These improvements make ADEPT highly suitable for real-time smart grid deployment while maintaining interpretability for operational decision-making.