Smart Hybrid Intelligence for Day-Ahead Energy Forecasting and Demand Response in Smart Grid

Accurate short-term energy forecasting and effective demand response (DR) coordination are critical for modern smart grids operating under high renewable penetration. This paper presents a hybrid metaheuristic–machine learning framework that integrates particle swarm optimization–tuned Long Short-Term Memory (LSTM) models for day-ahead renewable generation forecasting with a lightweight regression-based consumption predictor. The framework incorporates an imbalance-aware DR module that triggers selective load-shifting based on predicted generation–consumption deviations. Experimental evaluation using real-world time-series data demonstrates that the proposed method achieves day-ahead forecasting errors of 6.38 kWh for generation and 2.61 kWh for consumption, while the DR mechanism reduces net energy imbalance by approximately 23%. These results highlight the potential of hybrid intelligence approaches for enhancing grid flexibility, improving operational planning, and supporting sustainable energy management.