Reinforcement Learning Based Solar Power Nowcasting with Real-time Cloud Path Prediction

The integration of solar power into Sri Lanka’s power grid is challenged by the intermittent nature of solar energy due to cloud cover. This research presents an intelligent control system for real-time cloud path prediction and solar power forecasting. Utilizing a sky-imaging device with a fisheye lens and CMOS IMX179 sensor, the system captures high-resolution sky images. A custom YOLOv8 model detects clouds and the sun, while ByteTrack and centroid-based tracking analyze cloud trajectories. Reinforcement learning is employed to predict cloud trajectories, and a random forest algorithm forecasts power output fluctuations. Results demonstrate the system’s effectiveness in providing accurate, real-time solar power forecasts, contributing to enhanced grid stability and supporting renewable energy goals in Sri Lanka.