电工技术学报2025，Vol.40Issue(23)：7664-7676,13.DOI:10.19595/j.cnki.1000-6753.tces.242073
基于I-V曲线全局特征提取的光伏组串Swin-Transformer故障诊断方法
Fault Diagnosis Method for Photovoltaic String Based on Global I-V Curve Feature Extraction Using Swin Transformer
摘要
Abstract
The installed capacity of grid-connected photovoltaic power generation has grown rapidly,and the healthy operation of PV strings is crucial to the safety and stability of the PV power plant system.However,with the construction of PV bases in deserts,Gobi wastelands,and mountainous areas,the harsh outdoor operating environment has led to frequent failures,resulting in a decline in the overall power and even fire accidents.Therefore,fast and accurate PV fault diagnosis is crucial to ensure the reliable and safe operation of PV bases.In order to improve the automated operation and maintenance level of PV system,a swin-transformer fault diagnosis method for PV strings based on global feature extraction of I-V curves is proposed to realize accurate and reliable intelligent PV condition monitoring. Firstly,global I-V feature extraction is carried out to make full use of the dynamic characteristics of the I-V curve,which is greatly affected by the environment,and in order to avoid misidentifying the changes in the curve caused by environmental factors as fault characteristics,the curve needs to be calibrated to the standard test conditions(STC).After calibration,normalization preprocessing is performed due to the different sampling points and uneven distribution.The normality of the I-V curve data is improved by the correction and normalization preprocessing.Then,the dynamic characteristics of I-V curves are portrayed multidimensionally using the Gram angle field(GAF),recurrence plot(RP)and relative position matrix(RPM)feature transformations,and the global I-V features characterizing the state information of PV arrays are extracted. The GAF,RP and RPM feature extraction methods transform the data into images for processing and analysis using models in the field of computer vision,and for the characteristics of periodicity and repetitiveness of local areas of the image,a fault diagnosis model based on swin transformer is proposed,which adopts a hierarchical structure to progressively aggregate the local features,and then represents the global features,and the hierarchical feature extraction is more suitable for dealing with the image data;a shift window is designed to extract global I-V features characterizing the state information of PV arrays.The hierarchical feature extraction method is more suitable for processing image data;a shift window mechanism is designed to fuse local and global features across the window,which has strong image processing capability;local self-attention computation using the shift window can adapt to image inputs with different resolutions without providing a large number of samples,which effectively reduces the amount of computation and improves the efficiency of the model,and realizes high-precision fault diagnosis. Simulations and field experiments on a 3.75 kW PV system show that the proposed method performs best under the relative position matrix feature transformation,and can accurately diagnose multiple faults with different conditions and severities.The model accuracy is 99.67%with as low as 25 samples per class and 99.56%with 30dB noise interference.Ablation experiments using multiple feature data with different algorithms validate the superiority of the proposed feature extraction method and fault diagnosis model.This study provides reliable technical support for the stable operation of PV strings.关键词
光伏组串/故障诊断/I-V曲线/全局特征/Swin-TransformerKey words
PV strings/fault diagnosis/I-V curve/global feature/Swin-Transformer分类
信息技术与安全科学引用本文复制引用
昌千琳,罗永捷,王强钢,任博,周念成..基于I-V曲线全局特征提取的光伏组串Swin-Transformer故障诊断方法[J].电工技术学报,2025,40(23):7664-7676,13.基金项目
国家自然科学基金集成项目资助(U22B6006). （U22B6006）
