Machine Learning-Based Prediction of Solar Energy Output for Green Power Management

Accurate prediction of solar energy output is critical for efficient green power management in smart grid systems. This paper proposes a hybrid machine learning framework combining Long Short-Term Memory (LSTM) networks, Random Forest (RF), and Gradient Boosting (GB) regression for forecasting photovoltaic (PV) solar energy generation. The proposed ensemble model is trained on a multi-year dataset comprising meteorological variables including irradiance, temperature, humidity, and cloud cover. Experimental results demonstrate that the hybrid LSTM-RF-GB model achieves a Mean Absolute Percentage Error (MAPE) of 3.21% and an R² score of 0.974, outperforming conventional methods by 18–27%. The system is validated on real-world data from a 5 MW solar farm in Rajasthan, India, and further evaluated across multiple climate zones. Economic analysis demonstrates potential annual savings of ₹42.6 lakh through optimized grid dispatch enabled by accurate forecasting.