基于多时相巡检图像的变电设备缺陷检测仿真  

Simulation of Substation Equipment Defect Detection Based on Multi-temporal Inspection Image

作　　者：程骋 李科 黄海 Chen Cheng;LI Ke;Huang Hai(Nanchang Power Supply Branch of State Grid Jiangxi Electric Power Co.,Ltd.,Nanchang City,Jiangxi Province 330200;Yichun Power Supply Branch of State Grid Jiangxi Electric Power Co.,Ltd.,Yichun City,Jiangxi Province 336000;Tonggu County Power Supply Branchof State Grid Jiangxi Electric Power Co.,Ltd.,Yichun City,Jiangxi Province 336200)

机构地区：[1]国网江西省电力有限公司南昌供电分公司,江西南昌330200 [2]国网江西省电力有限公司宜春供电分公司,江西宜春336000 [3]国网江西省电力有限公司铜鼓县供电分公司,江西宜春336200

出　　处：《计算机仿真》2025年第2期447-452,共6页Computer Simulation

摘　　要：变电设备智能化巡检能有效提升电力网运营成本、提升电网运行效率。为解决当下计算机视觉只能巡检中存在缺陷检测稳定性差、泛化能力弱的问题,将Cycle GAN图像迁移算法与改进YOLOv5缺陷检测算法融合,通过扩充多时相巡检图像数据集,提升变电设备缺陷检测模型的鲁棒性。首先采用像素加权平均灰度化与邻域中值替代滤波处理,提升图像风格可迁移的适应性;然后利用Cycle GAN网络,在优化损失函数的基础上,基于S2W季节数据集优化迁移扩充PISE变电设备巡检数据集,提高数据图像的多时相性能;接着引入了可插入式CBAM结构块和自适应锚点框结构,优化传统的YOLOv5网络,提升微型变电设备缺陷检测性能;最后基于S2W-PISE多时相变电设备巡检数据集,训练构建改进YOLOv5变电设备缺陷检测模型。多时相变电设备缺陷检测仿真结果显示,与R-CNN、NMS和SSD三类传统变电设备缺陷检测模型相比,在S2W-PISE数据集与缺陷检测评价系统中,改进YOLOv5模型的P参数指标平均提升了5.79%,R参数指标平均提升了5.80%,即具有最高的缺陷检测精度与稳度,同时改进模型具有最优的综合性与鲁棒性。综上所述,改进YOLOv5变电设备巡检图像缺陷检测算法在电网智能化发展中具有重要的仿真价值。Intelligent inspection of substation equipment can effectively improve the operation cost and efficiency of power grid.In order to solve the problems of poor stability and weak generalization ability of defect detection in the current computer vision inspection,this paper integrates the Cycle GAN image migration algorithm with the improved YOLOv5 defect detection algorithm,and improves the robustness of the substation equipment defect detection model by expanding the multi-temporal inspection image data set.Firstly,pixel weighted average graying and neighborhood median substitution filtering are used to improve the adaptability of image style migration.Then,on the basis of optimizing the loss function,Cycle GAN network is used to expand the PISE substation equipment inspection data set based on S2W seasonal data set optimization migration to improve the multi-temporal performance of data images;Then,the traditional YOLOv5 network is optimized by introducing the pluggable CBAM structure block and the adaptive anchor box structure to improve the defect detection performance of micro-substation equipment.Finally,the improved YOLOv5 substation equipment defect detection model is trained and constructed based on the S2W-PISE multi-phase substation equipment inspection data set.The simulation results show that,compared with R-CNN,NMS and SSD,the P parameter index of the improved YOLOv5 model increases by 5.79%and the R parameter index increases by 5.80%on average in the S2W-PISE data set and defect detection evaluation system.That is to say,it has the highest accuracy and stability of defect detection,and the model has the best comprehensiveness and robustness.To sum up,the improved YOLOv5 image defect detection algorithm for substation equipment inspection has important simulation value in the development of intelligent power grid.

关 键 词：变电设备多时相图像 缺陷检测 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]