Research And Implementation Of Fault Component Identification System For Transmission Line

With the continuous expansion of the scale of the power system,the safety of transmission lines and their facilities is crucial to the normal operation of the power system,and insulators and anti-vibration hammers in transmission lines are very critical components,and are also the key inspection objects in the inspection tasks.The role of anti-vibration hammer is to prevent violent vibration of the wire under the influence of strong wind,but due to its long-term contact with air in the outside world,resulting in its easy to be corroded off.The insulator is an insulating device on the transmission line,which is not only subject to general mechanical and electrical loads,but also highly susceptible to erosion from the external environment and prone to self-detonation.These faults will bring great threat to the normal operation of transmission lines.Therefore,in the task of transmission line inspection,insulator and anti-vibration hammer fault detection becomes one of the important links.This paper focuses on the detection of anti-vibration hammer shedding and insulator self-detonation faults in transmission line UAV aerial images.First,by analyzing the characteristics of insulator self-detonation and anti-vibration hammer dislodgement,the corresponding datasets is constructed.Then,Cascade RCNN(Cascade Region Convolutional Neural Networks)is used as the base network for fault detection,and the detection study is conducted for anti-vibration hammer dislodgement and insulator selfdetonation to improve the accuracy of fault detection.The main contributions of this paper are as follows:(1)In order to improve the ability to distinguish between positive and negative samples during network training,a contrast learning network is proposed.The Euclidean distance is used to calculate the feature similarity between positive and negative samples output by the region proposal network and the real samples respectively,and then the feature similarity is input into the contrast loss function to train the network,so as to improve the differentiation between positive and negative samples.The network alleviates the problem of obscured target features extraction not obvious and improves the model detection ability.The experimental results show that compared with the baseline model,the recall of the improved model is improved by 5.5%to 95.6%,and the average precision is improved by 4.8%to 88.9%.(2)In order to improve the classification performance of the model classifier,a classifier enhancement method is proposed,by designing additional branches on the cascade structure of the baseline network to filter out the region of interest features with better regression effects in the cascade structure and input them directly into the final classification regression queue.This improvement alleviates the problem that some regions of interest become poorly regressed after the cascade structure,optimizes the shared features of the classification-regression task,and thus improves the classification performance of the classifier.The average accuracy of the improved model is further improved by 1.4%to 90.3%after testing and analysis.(3)To address the problem of feature confusion caused by the close proximity of dislodged antivibration hammers and normal antivibration hammers,a parallel attention mechanism module is designed,which mainly consists of channel attention and spatial attention in a parallel structure.By reintegrating the features of the backbone network input,the module increases the weights of key features and effectively improves the feature extraction capability of the model,thus alleviating the problem of inaccurate network localization when the dislodged antivibration hammer is close to the normal antivibration hammer.After testing and analysis,the average accuracy of the improved model is further improved by 1.6%,reaching 91.9%.(4)To address the problem of poor detection of self-explosion targets at the joints between the pegboard and insulator string in the insulator data set,a self-explosion detection network for insulators that incorporates multi-scale contextual information is designed to enhance the feature association between the target and the surrounding environment during the network training.The network is able to better learn the association between target features and contextual information and make full use of the contextual features of the target,thus alleviating the problem of confusing this type of self-exploding defects with normal insulator pieces.Experimental results show that the average accuracy of the improved model is improved by 6.2%to 91.2%relative to the baseline model.(5)A transmission line fault component identification system is developed,with the following module designs:system login module,user management module,data upload module,fault component detection module,and result visualization and statistics module.In order to meet the practical requirements,a detailed design is provided,in terms of requirement analysis,functional module design,data relationship model and system architecture,and the application of transmission line fault component identification methods is explored.In this paper,improvement methods are proposed for the identification difficulties existing in the anti-vibration hammer and insulator datasets respectively,and the performance is improved to a certain extent.The improved model for anti-vibration hammer dislodgement detection achieves an average accuracy of 91.9%and a recall rate of 98.6%,which are 7.8%and 8.5%higher than the baseline model,respectively;the improved model for insulator self-detonation achieves an average accuracy of 91.2%and a recall rate of 96.9%,which are 6.2%and 6.8%higher than the baseline model,respectively.The method proposed in this paper improves the recognition accuracy of the above two fault components and can help improve the level of intelligent inspection of transmission lines.