Research on short-term prediction of photovoltaic power via improved VMD-based dynamic model fusion and adaptive boundary optimization

Abstract Probabilistic prediction of photovoltaic (PV) output power is crucial to maintain the stable operation and reliability of the power grid and to develop effective operational strategies and short-term plans. Therefore, a new short-term PV power generation forecasting approach via improved variational mode decomposition (I-VMD)-based dynamic model fusion and adaptive boundary optimization is proposed in this paper. First of all, a correlation analysis is conducted on various meteorological data that affect PV output. Then, feature extraction of selected meteorological data is carried out by using I-VMD and kernel principal component analysis (KPCA). After that, based on the improved stacked generalization (I-Stacking) ensemble learning framework, a new multi-model fusion method is proposed to construct a forecasting model for short-term PV power generation. Finally, an adaptive boundary optimization method for prediction errors is proposed to enhance the PIs overall performance. Through numerical comparison and analysis with the conventional methods, the performance of the proposed method is validated.