Accurate minute-ahead solar-power forecasts are essential for grid stability and market bidding in India’s rapidly expanding photovoltaic sector. We develop a time-aware Random Forest Regression (RFR) framework that ingests synchronized plant-level power, weather-sensor, and temporal signals. The merged Plant 1 Generation and Weather Sensor datasets span 34 consecutive days (15 May–17 June 2020) and yield 68 774 one-minute records after alignment. Cyclical time features (hour, day, month) and solar-elevation proxies expose diurnal–seasonal regularities, while Seasonal–Trend decomposition using Loess (STL) isolates slowly varying irradiance trends from high-frequency cloud-induced fluctuations to improve model explainability. The ensemble achieves computational efficiency through parallel tree construction, enabling deployment on low-resource edge devices. Across diverse climatic conditions, the tuned model attains RMSE = 0.00048, MAPE = 0.069%, and R2 = 0.99996, underscoring its reliability for grid-level decision making. Feature-importance analysis shows that DC_POWER contributes 83% of predictive influence, with irradiation and module temperature providing additional explanatory power, ensuring decisions remain interpretable for operators. By uniting lightweight computation, transparent reasoning, and sub-percent error rates, the proposed time-aware RFR offers a practical, field-ready alternative to deep-learning models for real-time AC-power forecasting and paves the way for hybrid physical–AI extensions under extreme weather scenarios.