基于贝叶斯优化集成多种光流法的云图预测技术OA

Cloud variation significantly influences Photo Voltaic(PV)power generation,and cloud map prediction using optical flow methods often suffers from instability when spatio-temporal scenario changes.In this study,a cloud map prediction technique that leverages Bayesian optimization to integrate multiple optical flow methods was proposed.Specifically,multiple optical flow methods are independently applied to process cloud map sequences and extract cloud motion vectors.The accuracy of each method's historical predictions is then evaluated based on near-real-time observations,allowing Bayesian optimization to determine the optimal weights for integrating these methods.Experiments conducted in the middle and lower reaches of the Yangtze River demonstrate that the proposed integration method improves cloud map prediction accuracy and long-term stability compared to multi-model averaging approaches.This method also excels in preserving image fidelity and structural similarity.Furthermore,it effectively mitigates the performance degradation typically associated with extended prediction time spans and exhibits enhanced stability in multi-step prediction.By advancing cloud map prediction accuracy,this study provides a valuable contribution to PV power generation forecasting and solar resource utilization under cloudy and rainy weather conditions.