An Explainable AI Framework for Net-Load Forecasting and Optimization-Based Automatic Generation Control in Renewable-Rich Indian Power Grids

The growing penetration of solar and wind generation in India has significantly increased net-load variability, making accurate anticipation of rapid fluctuations and effective frequency regulation increasingly challenging for system operators. Conventional forecasting approaches often treat demand and renewable generation separately and offer limited interpretability, reducing operator confidence and hindering informed stability planning. This study proposes an integrated explainable artificial intelligence (XAI) framework that links satellite-based weather intelligence, machine-learning–driven net-load forecasting, and optimization-based automatic generation control (AGC) analysis for renewable-rich Indian power grids. A merged daily dataset for 2025 was constructed using NLDC–REMC operational reports and NASA POWER/ERA5 meteorological variables, yielding 136 samples with 35 grid and weather features. After feature engineering, a 52-variable supervised matrix was developed for day-ahead net-load forecasting. Among the evaluated models, the XGBoost regressor demonstrated the most reliable performance, achieving an RMSE of 225.69 MW, MAE of 189.04 MW, and MAPE of 6.37%, outperforming LSTM, BiLSTM, Random Forest, ARIMA, and persistence baselines. SHAP-based explainability revealed that short-term temporal memory, net-load ramp indicators, and meteorological drivers such as humidity and irradiance dominate predictive accuracy, ensuring physically meaningful and transparent model behavior. Building upon the forecasting stage, the framework advances to real-time grid operation by formulating an AGC problem for a two-area reheat thermal power system. Proportional–integral controller gains are optimally tuned using Artificial Lemming Algorithm (ALA), Particle Swarm Optimization (PSO), and Genetic Algorithm (GA). Time-domain simulations demonstrate substantial reductions in peak frequency deviation, settling time, tie-line power oscillations, and integral error indices compared to the baseline controller. Robustness studies under varying load disturbances and an inertia-proxy analysis further highlight the importance of optimized AGC tuning in low-inertia, renewable-dominated grids. Overall, the proposed framework offers a coherent, interpretable, and operationally relevant decision-support tool for enhancing frequency security as India progresses toward deep renewable-energy integration.