| Ensuring the safe operation of overhead transmission lines is one of the important part of the power grid operation and maintenance.External damage is one of the important causes of overhead line outages,but due to the wide distribution of overhead lines,it is extremely difficult to rely on human supervision and prevention.Therefore,it is necessary to adopt safety monitoring means based on its vision for overhead transmission lines to prevent the occurrence of external damage.However,the recognition accuracy,alarm accuracy and early warning timeliness of the on-line monitoring system based on machine vision still can not meet the requirements of engineering application.To this end,this paper carries out the research of online monitoring technology for external force damage of overhead transmission lines based on machine vision.It includes the construction of overhead transmission line external force damage sample library,the research of recognition algorithm based on deep learning and the design of monitoring system with integrated front and rear end recognition,in order to realize real-time monitoring and target detection of external force damage abnormal targets.This paper mainly accomplishes the following research work.(1)The deep learning algorithm requires a large number of samples for training and there is no publicly available sample library of overhead transmission line external damage.In this paper,the sample expansion method is studied on the premise that the number of original overhead line external damage images is small.The sample expansion is achieved by two methods based on perturbation and image synthesis,and finally,the sample library of overhead transmission line external force damage images is established by manual annotation with Label Img,which is suitable for deep learning training.(2)Combining the advantages of fast detection speed and small memory consumption of SSD and fast convolution speed in Mobile Net’s deep separable convolution,a transmission line external damage anomaly target detection algorithm for front-end identification is proposed.On the basis of Mobile Net-SSD,feature extraction network structure adjustment and feature fusion are made to form a new target detection network ILSM-SSD.Through model training and testing,the detection accuracy of the algorithm in this paper is significantly improved on the front-end embedded hardware platform.Therefore,the application effect of the identification of the front-end of the transmission line damage caused by external forces is improved.(3)In order to solve the problem that the performance of the hardware platform with the deep learning algorithm in the front end is not good enough to improve the accuracy of the algorithm recognition,this paper proposes to carry out the second recognition in the main station of the back end and realize the alarm through the front and back end comprehensive recognition.After the system is designed and implemented,it is applied in a power supply bureau.After running for months,the recognition rate of transmission line on-line monitoring system based on front and back end recognition is over 95%.The background master station can improve the situation of misjudgment and missing judgment in front-end identification,and effectively improve the speed and accuracy of detection.In summary,this paper,by studying three engineering application improvement methods of the overhead line external damage online monitoring system,significantly improves the accuracy of external damage identification and the timeliness of dangerous target warning.On the one hand,it improves the practicality of the current transmission line online monitoring system,and on the other hand,it follows the current development trend of cloud-side collaboration and edge computing technology to solve practical engineering problems,which plays an important role in ensuring safe line operation. |
PDF Full Text Request