A EOF-DBSCAN Hybrid Model for Distributed Photovoltaic Cluster Power Prediction

Accurate prediction of distributed photovoltaic (PV) cluster output is crucial for ensuring grid integration security and improving the operational efficiency of PV power stations. This paper integrated with EOF-DBSCAN clustering and Dynamic Time Warping (DTW) similarity day selection to optimize data preprocessing and cluster partitioning. To tackle challenges including complex feature correlations among power stations and significant deviations in meteorological time series similarity measurement, the EOF-DBSCAN method is employed to integrate station correlations and geographical features, achieving refined cluster division. The DTW algorithm is utilized to accurately select similar days and optimize weight time series. In comparative experiments, traditional forecasting models such as Linear Regression and XGBoost were eclipsed by the proposed method, which achieved higher accuracy consistently under a range of meteorological conditions. and LSTM.