基于Voronoi图与改进TT-GAN的光伏时序数据增强方法OA

The scarcity of measured data from photovoltaic(PV)power stations and data gaps caused by sensor fail-ures or communication interruptions significantly compromise the accuracy and robustness of power forecasting.To address these challenges,this paper proposes a novel data enhancement method for PV time-series data based on Voronoi diagram and transfer learning-enabled transformer-based generative adversarial network(TT-GAN).The method aims to establish a unified mechanism for multi-source domain selection and data enhancement framework.A Voronoi diagram combined with a data-strength metric is used to dynamically select source station sets under di-verse scenarios.Subsequently,a TT-GAN model is developed,incorporating structural and fine-tuning optimiza-tions.Enhanced with a self-attention mechanism and supervised training,the model improves its ability to handle data noise and feature representation learning,making it suitable for both data generation and data repair tasks.Ex-perimental results demonstrate that the proposed model significantly improves data quality on existing PV time-series datasets and enhances the accuracy of power forecasting.