Masood Ur Rehman: Context–Aware Solar PV Forecasting for Critical Energy Infrastructure using Seasonal Mixture of Experts

This paper proposes a seasonal Mixture of Experts (MoE) framework that integrates season-specialized Long Short-Term Memory (LSTM) experts with a context-aware gating mechanism to improve forecasting accuracy and adaptability.

Each expert captures the temporal characteristics of a specific season, while the gating network dynamically weights their contributions based on the current operating context. Qualitative analysis of expert selection confirms the model’s adaptability to evolving seasonal dynamics, demonstrating the effectiveness of combining seasonal specialization with adaptive expert blending for accurate and generalizable solar PV power forecasting.