Distributed PV Forecasting Technology Based on Data Enhancement and Lightweight Model Dynamic Correction Method

At present, short-term power prediction of distributed new energy sources is confronted with problems such as data deficiency, and new methods need to be developed. To address this issue, short-term power prediction of grid-based distributed photovoltaic power station clusters can be utilized. Through processes such as grid preparation, historical and irradiation data augmentation, grid-based power prediction, and effect evaluation, missing power data can be reconstructed using data from surrounding stations and grid-based prediction can be conducted. Meanwhile, for the incorporation of multi-source data, a dynamic correction technology for lightweight models of distributed new energy power prediction can be adopted. By applying knowledge distillation and incremental learning, multisource data can be used to dynamically correct the grid prediction results, thereby enhancing prediction accuracy.