摘要
针对光伏组件的热斑和蒙尘故障,提出基于YOLOv7的故障检测模型。首先,引入Mosaic与Mixup结合的数据增强方法扩充图像数据集,提高模型的泛化能力;其次,引入坐标注意力机制(CA),不仅能关注通道特征和空间特征,还能解决长程依赖的问题。实验结果表明,改进后的YOLOv7模型在检测光伏组件、蒙尘以及热斑故障时的准确率分别达到96.08%、83.92%以及77.19%,与原始模型相比分别提升3.66个百分点、1.90个百分点以及2.27个百分点;mAP值由82.48%提升至83.05%;模型精度提高,鲁棒性增强,满足实际应用需求。
Aiming at the hot spot and dust fault of photovoltaic modules,this paper proposes a fault detection model based on YOLOv7.Firstly,a Data Augmentation method combining Mosaic and Mixup is introduced to expand the image dataset and enhance the model′s generalization ability.Secondly,the Coordinate Attention mechanism is incorporated to effectively focus on channel features and spatial features while addressing long-range dependence issues.The experimental results show that the accuracy of the improved YOLOv7 model in detecting photovoltaic modules,dust faults and hot spot faults reaches 96.08%,83.92%and 77.19%respectively,which is 3.66 percentage points,1.90 percentage points and 2.27 percentage points higher than that of the original model.The mean average precision(mAP)increases from 82.48%to 83.05%.The accuracy of the model is improved and the robustness is enhanced to meet the needs of practical applications.
作者
张文馨
周宇
王劲松
李忠艳
Zhang Wenxin;Zhou Yu;Wang Jinsong;Li Zhongyan(School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China;School of Mathematics and Physics,North China Electric Power University,Beijing 102206,China;North China Electric Power Test and Research Institute,China Datang Corporation Science and Technology General Research Institute Co.,Ltd.,Beijing 100043,China)
出处
《太阳能学报》
北大核心
2025年第8期333-340,共8页
Acta Energiae Solaris Sinica
基金
国家重点研发计划(2020YFB1707802)
国家自然科学基金(12071131)。
关键词
光伏组件
目标检测
深度学习
卷积神经网络
图像处理
PV modules
object detection
deep learning
convolutional neural network
image processing
作者简介
通信作者:李忠艳(1972-),女,博士、教授,主要从事泛函分析、框架小波分析、机器学习等领域的应用研究。lzhongy@ncepu.edu.cn。
