基于改进YOLOv8的输电线绝缘子故障诊断方法  

Fault diagnosis method for transmission line insulator based on improved YOLOv8

作　　者：龚闯 江维 邹德华 于俊康[1] GONG Chuang;JIANG Wei;ZOU Dehua;YU Junkang(School of Mechanical Engineering and Automation,Wuhan Textile University,Wuhan 430073,China;Hunan Province Key Laboratory of Intelligent Live Working Technology and Equipment(ROBOT)(State Grid Hunan Ultra-High Voltage Transmission Company),Changsha 420100,China;Live Inspection and Intelligent Operation Technology State Grid Corporation Laboratory(State Grid Hunan Ultra-High Voltage Transmission Company),Changsha 420100,China)

机构地区：[1]武汉纺织大学机械工程与自动化学院,湖北武汉430073 [2]智能带电作业技术及装备(机器人)湖南省重点实验室(国网湖南省电力有限公司超高压输电公司),湖南长沙420100 [3]带电巡检与智能作业技术国家电网公司实验室(国网湖南省电力有限公司超高压输电公司),湖南长沙420100

出　　处：《河南工程学院学报(自然科学版)》2025年第1期38-44,共7页Journal of Henan University of Engineering:Natural Science Edition

基　　金：湖南省重点实验室开放性课题资助项目(2024KZD1001)。

摘　　要：输电线路绝缘子故障主要包括污闪、自爆、破裂等,不同类型故障有相应的检修方式,故提出了一种基于改进YOLOv8的绝缘子故障诊断方法。首先,通过数据增强方法扩充数据集,为减少模型的参数量并提高计算效率,引入VanillaNet轻量化YOLOv8的原有主干网络。然后,在模型中集成AFPN结构,提高模型对不同目标的检测能力,并引入BiFormer双层级注意力机制加强目标的特征提取,增强模型的检测性能。最后,进行实验验证并得出结论,引入的AFPN结构使模型尺寸缩小了13.1%,有助于在资源受限的设备上部署,引入VanillaNet结构使得模型尺寸缩小了49.82%,引入BiFormer注意力机制,在不明显增大模型尺寸的前提下,提升了特征表示的质量。对模型进行实际检测任务的验证,结果表明改进模型检测效果良好,检测速度相较于传统模型提升了20.8%。Insulator faults of transmission lines mainly include pollution flashover,self-explosion,rupture,etc.Different types of insulator faults have their own maintenance methods.Therefore,this paper proposes an improved YOLOv8 of insulator fault diagnosis method.Firstly,the dataset is expanded through data augmentation methods.In order to reduce the number of model parameters and improve computational efficiency,the original backbone network of VanillaNet lightweight YOLOv8 is introduced.Then,an AFPN structure was integrated into the model,which can effectively fuse multi-scale features and improve the model′s detection ability for targets of different sizes,a BiFormer two-level attention mechanism is introduced to enhance the feature extraction of the target and enhance the detection performance of the model.Finally,based on the experimental results,the introduction of the AFPN model reduced the model size by 13.1%,which is beneficial for deployment on resource constrained devices.On this basis,the introduction of the VanillaNet structural model reduced the model size by 49.82%,and the introduction of the BiFormer attention mechanism improved the quality of feature representation without significantly increasing the model size.The model was validated for practical detection tasks,and the results showed that the improved model had good detection performance,with a detection speed improvement of 20.8%compared to traditional models.

关 键 词：绝缘子故障诊断 YOLOv8 数据增强 轻量化网络 注意力机制 

分 类 号：O232[理学—运筹学与控制论]