| Overhead transmission line is an important guarantee for the stable operation of power grid,which is closely related to the development of national economy and social stability.It is very important to ensure that the transmission line is not damaged by external forces.In recent years,with the continuous expansion of the scale of China’s cities and power grids,the incidence of power security accidents is also gradually increasing.According to the statistics of domestic power safety accident analysis data,about one third of power safety accidents are caused by external forces.At present,the detection of dangerous behavior of transmission line damage by external force is mainly carried out by means of regular manual line inspection,UAV line inspection,sensor and real-time monitoring.Manual regular line patrol needs a lot of manpower,material resources and time,and the work efficiency is low.Compared with manual line patrol,UAV improves the efficiency of line patrol,but the working distance and cruise time of UAV are limited,so it is difficult to be widely used in power system.Although the sensor detection method is not limited by distance and time,there are still some technical problems: the accuracy of the detector is low,it is vulnerable to bad weather,it is unable to accurately determine the type of external force damage dangerous behavior in the warning area,and the damage degree can only be determined by manual on-site investigation.The real-time monitoring mode requires the staff to identify the types of dangerous behaviors caused by external forces,and the detection efficiency and accuracy can’t be guaranteed.The improvement of real-time monitoring coverage produces a large amount of image data,and the processing of these data needs the help of human participation,which increases the workload and reduces the work efficiency.The way of real-time monitoring does not fundamentally realize the intelligent detection of the dangerous behavior of transmission line damage caused by external force.It is expected that through the use of electronic image data analysis,pattern recognition,machine deep learning and other technical means,the video content of power monitoring line can be automatically interpreted in real time,and the dangerous behaviors and types of dangerous behaviors that may occur in the warning area of power transmission line can be detected and identified in real time,In this way,it can effectively reduce a lot of human resources and processing time required by using the traditional line monitoring video method.In this paper,based on deep learning and target detection technology,a method of transmission line external force damage dangerous behavior detection is proposed.The main idea of this paper is:first of all,the research status at home and abroad and the necessity of this study are described.Then,it introduces the technical means of the research: convolution neural network and target detection technology.By comparing the principles of several target detection models,yolov3 algorithm is selected as the basic algorithm of dangerous behavior detection.Secondly,the data set is constructed.Firstly,the data is expanded by using data enhancement,and the expanded data is annotated.Finally,according to the characteristics of the research object in this paper,the algorithm based on the selection is improved.The main improvement points are as follows: 1.Network structure improvement: two convolution layers are added to the original network and one detection scale of the original network is deleted.The image data in the test set is used to test and verify the rationality and effectiveness of the improvement.2.Improvement of loss function: reference giou and focal loss algorithm to improve the loss function of yolov3,and test the image data in the test set to verify the rationality and effectiveness of the improvement.3.Key technology improvement in the algorithm: refer to K-means + + algorithm and soft NMS algorithm to improve the clustering analysis and non-maximum suppression algorithm in yolov3.The image data in the test set is used to test the rationality and effectiveness of the improvement.Finally,the image data set after data enhancement processing in this paper is used to verify.The data show that the detection performance and speed of the improved algorithm and the original yolov3 detection algorithm have been greatly improved,and the detection accuracy map(mean average precision)has been improved by 4.2%,and the accuracy has been improved by 5.2%. |
