An improved cross-correlation method for efficient clouds forecasting

Accurate real-time forecasting of cloud positions relative to the sun is essential for photovoltaic (PV) scheduling. Due to its simplicity, the Cross-Correlation Method (CCM) has been widely used in short-term solar radiation prediction by estimating Cloud Motion Vectors (CMVs). However, comparing each pixel block in the current image with neighboring rectangular areas in the previous image results in increased computational complexity for CCM, which limits its application in short-term solar radiation forecasting. This study utilizes the parallel optimization capability of the Particle Swarm Optimization (PSO) algorithm to simultaneously search for the best matching block for multiple pixel blocks to improve CCM's usability. The results show that the proposed PSO-CCM can estimate CMVs at least five times faster while ensuring forecasting accuracy, significantly reducing the computational complexity of traditional CCM. In addition, as a crucial step in cloud image forecasting, the extrapolation strategy of cloud images also plays a crucial role. Thus, this study also explores and analyzes in detail the advantages and disadvantages of two extrapolation strategies: Block-Wise Forecasting Method and Frozen Cloud Advection Method, offering valuable insights for short-term solar radiation forecasting models based on ground-based sky images.