摘要
对气体绝缘开关设备(gas insulated switchgear, GIS)典型部件的目标识别和温度提取是实现对设备发热状态红外智能检测的关键。文中提出一种基于混合域注意力机制(convolutional block attention module, CBAM)的改进YOLOv4算法,可实现对GIS母线、隔离开关等部件的快速目标检测和热点温度提取。首先,在某变电站现场采集原始红外图像,对图像进行锐化处理和部位标记,构建包含GIS典型部件的红外数据集。然后,利用深度可分离卷积网络降低模型参数量,并融入CBAM优化模型的识别能力,在此基础上构建基于改进YOLOv4的GIS红外部件目标快速检测算法。最后,采用灰阶差值方法对检测到的GIS典型目标部件进行热区温度值提取。结果表明,所提算法在GIS红外特征数据集上可以达到每秒31.5帧的识别速度和82.3%的识别准确率,明显优于其他目标算法,且GIS各部件的温升计算值与实测值误差在±1℃内。该算法可部署在无人机和巡检小车等边缘智能终端,实现对现场GIS设备温升状态的精细化识别和快速诊断,提升GIS设备健康状态管理数字化和智能化水平。
Target recognition and temperature extraction of the typical component of gas insulated switchgear(GIS) are the key to realizing the infrared intelligent detection of equipment heating state. In this paper, an improved YOLOv4 algorithm based on convolutional block attention module(CBAM) is proposed to achieve rapid target detection and hot spot temperature extraction of GIS bus, disconnector and other components. Firstly, the original infrared images are acquired at a substation site, and an infrared dataset containing typical GIS components is constructed by sharpening the images and marking components. Then, the deep separable convolutional network is used to reduce the amount of model parameters, and the CBAM is integrated to optimize the recognition ability of the model, on the basis of which a GIS infrared component target rapid detection algorithm with improved YOLOv4 is constructed. Finally, the gray-scale difference method is used to extract the temperature value of the hot area for the detected typical target components of GIS. The results show that the proposed algorithm can achieve a recognition speed of 31.5 frame per second and an recognition accuracy of 82.3% on the GIS infrared feature dataset, which is significantly better than other target algorithms. The error between the calculated value and the measured value of temperature rise of GIS components is within ±1 ℃. The algorithm proposed in this paper can be deployed in edge intelligent terminals such as unmanned aerial vehicles and inspection trolleys to achieve refined identification and rapid diagnosis of the temperature rise status of on-site GIS equipment, thus improving the digitalization and intelligence level of health management of GIS.
作者
刘江
关向雨
温跃泉
吕朝伟
LIU Jiang;GUAN Xiangyu;WEN Yuequan;LYU Chaowei(College of Electrical Engineering and Automation,Fuzhou University,Fuzhou 350108,China;State Grid Ganzhou Power Supply Company of Jiangxi Electric Power Co.,Ltd.,Ganzhou 341000,China)
出处
《电力工程技术》
北大核心
2023年第1期162-168,共7页
Electric Power Engineering Technology
基金
福建省自然科学基金资助项目(2020J01509)。
