Nonparametric Probabilistic Prediction of Regional PV Outputs Based on Granule-based Clustering and Direct Optimization Programming

Regional photovoltaic (PV) power prediction plays an important role in system planning and operation. To effectively improve the performance of intervals (PIs) for very short-term regional PV outputs, efficient nonparametric probabilistic method based on granulebased clustering (GC) direct optimization programming (DOP) is proposed. First, GC proposed to formulate cluster sample granules consisting numerical weather (NWP) historical output data, enhanced hierarchical performance. Then, accuracy samples' utilization, unbalanced extension used reconstruct training samples time series. After that, DOP applied quantify weights optimal overall Meanwhile, a balance coefficient studied reliability PIs. Finally, validated through multistep PIs comparison real generation data.