深度生成数字孪生框架下的光伏阵列故障诊断OA

Aimed at the problems of poor feature extraction and suboptimal diagnostic performance under small-sample scenarios,which are caused by difficulty in obtaining the photovoltaic(PV)array fault data under real operating condi-tions,a deep generative digital twins-based PV array fault diagnosis method is proposed.First,a mechanism model is constructed using the parameters of an actual PV system to generate typical fault data,addressing the issue of data scar-city.Second,a recursive feature elimination in cross validation module based on least squares support vector machine is developed to automatically evaluate and select the key features,enhancing the features'effectiveness.Third,a one-dimensional Pearson correlation coefficient UNet diffusion model is proposed to learn the intrinsic distribution character-istics of fault data and enhance the data diversity.Finally,a stochastic configuration network optimized by the aquila op-timizer is introduced to achieve an accurate fault diagnosis based on the augmented data.The proposed method was ex-perimentally validated on a 250 kW grid-connected PV system,achieving a fault diagnosis accuracy of 97.9%.