摘要
输电线路稳定运行是保障电力系统安全的重要环节之一,经典的机器学习算法对输电线路部件识别与分类准确率和效率都比较低。针对这一问题,选取了具有识别与分类功能的区域卷积神经网络(Faster-RCNN)来对部件进行识别与分类,研究了不同网络模型在输电线路中对不同部件的识别准确率和识别时间,结合实验结果,根据识别准确率和识别时间的优劣选取最佳网络模型,然后就如何提高模型的识别准确率和缩短识别时间展开研究,提出两种方法:通过调整CNN模型的卷积核大小和图像的旋转变换扩充数据集,实验结果表明两种方法都能有效的提高了输电线路巡检中的部件识别与缺陷检测的有效性和可靠性。利用无人机实际采集的图像进行识别和分类实验,实验结果表明深度学习方法在高压输电线路部件的识别与缺陷检测中的有效性和可靠性都非常高,Faster R-CNN进行部件识别与缺陷检测可以达到每张近0.17s的识别速度,对均压环的识别率最高可达到96.8%,mAP最高可以达到93.72%。
The stable operation of the transmission line is one of the important parts to ensure the safety of the power system.The classical machine learning algorithm has low accuracy and efficiency in the recognition and classification of transmission line components.Aiming at this problem,this paper selects the Region-based convolutional neural network(Faster-RCNN)to identify and classify the components,different network model in the transmission line is studied in the different components of recognition accuracy and recognition time,combined with the experimental results,according to the merits of the recognition accuracy and recognition time to select the best network model,and then how to improve the accuracy of the model recognition and shorten the recognition time to start research,this paper proposed two methods-by adjusting the kernel size of CNN model and image rotation transformation to expand the data set,the experimental results show that both methods can effectively improve the validity and reliability of component recognition and defect detection in transmission line inspection.Using the images by Unmanned Aerial Vehicle(UAV)collected to conduct recognition and classification experiment,the experimental results show that the validity and reliability of deep learning approach in identifying and defect detection of high voltage transmission line components is very high,and Faster R-CNN is used to identify and defect the parts,The detection rate can reach the recognition rate of nearly 0.17 s,the recognition rate of the equalization ring can be up to 96.8%,and the maximum mAP can reach 93.72%.
作者
汤踊
韩军
魏文力
丁建
彭新俊
Tang Yong;Han Jun;Wei Wenli;Ding Jian;Peng Xinjun(School of Communication and Information Engineering, Shanghai University, Shanghai 200444, China;State Grid Zhejiang Electric Power Company Maintenance Branch, Hangzhou 310007, China)
出处
《电子测量技术》
2018年第6期60-65,共6页
Electronic Measurement Technology
基金
国家自然科学基金(61471230)项目资助