复杂时空失配场景下分布式光伏功率鲁棒预测OA

To address the mismatch problem in sampling intervals and spatial locations between meteorological stations and photovoltaic panel monitoring,a robust forecasting framework for distributed photovoltaic power under complex spatiotemporal mismatch conditions is proposed.First,a mechanism-constrained spatiotemporal correction process is constructed to tackle the spatiotemporal mismatch between meteorological stations and photovoltaic panels.Through preprocessing and data calibration,a relative consistent mapping of meteorologi-cal observations to irradiance and temperature on photovoltaic panels is achieved,forming high-quality,tempo-rally coherent multi-source inputs.Secondly,a variational mode decomposition(VMD)module is employed to perform multi-scale decomposition of key meteorological and electrical variables.This is followed by a kernel principal component analysis(KPCA)module for nonlinear dimensionality reduction and redundancy suppres-sion,thereby enhancing feature representation and noise resistance.Finally,an iTransformer model is intro-duced as the temporal predictor to model the selected feature sequences,while the sparrow search algorithm(SSA)model is used to globally optimize critical hyperparameters,resulting in the VMD-KPCA-SSA-iTrans-former prediction model.The results show that under typical conditions such as sunny,cloudy,and rainy weather,the proposed model consistently outperforms comparison models in terms of key metrics such as the coefficient of determination(R2),root mean squared error(RMSE),and mean absolute error(MAE).