A robust hybrid approach combining cloud block matching and optical flow for photovoltaic forecasting

This paper presents a robust hybrid approach for cloud motion forecasting in intra-hour photovoltaic power generation forecasting. A gradient-based adaptive threshold approach is employed for cloud recognition, utilizing the Sobel gradient operator for initial segmentation and an adaptive normalized red-to-blue ratio for fine-tuning, effectively distinguishing clouds from the sky even in challenging scenarios. The proposed cloud motion prediction approach combines cloud block matching and optical flow to calculate cloud movements vectors, providing comprehensive information for accurate forecasting. Experimental results demonstrate the superior performance of the proposed approach compared to traditional approaches, helping to improve the accuracy and reliability of intra-hour photovoltaic power generation forecasting.