Research And Implementation Of Anomaly Detection Of Overhead Contact System

The catenary is an irreplaceable part of the high-speed railway,which is a power supply system built around the high-speed railway.Its function is to transmit electrical energy to high-speed trains through wires.The quality and working conditions of the catenary system have a direct impact on the transportation safety of high-speed railways.According to statistics,catenary faults account for 70% of railway faults.To ensure the safe operation of high-speed railway transportation,careful inspection of key catenary parts is required.The traditional manual inspection method has the shortcomings of low inspection accuracy and long inspection period.An effective inspection algorithm is needed to detect abnormal contact parts.Here,image processing and deep learning techniques are used to conduct specific research on the following aspects:(1)In view of the low detection accuracy of the existing catenary key parts detection algorithm,small catenary parts,poor lighting conditions and complex background,the paper proposes an improved YOLO detection algorithm.By increasing the length of the connection,the network Topological features and high-level features can be merged.When extracting features,the network can retain and refine small target features to increase sensitivity to small target detection;at the same time,a two-dimensional Dropout mechanism is added to solve the problem that the network is prone to overfitting.,Improve the generalization ability of the network;modify the category loss function of the network from the reverse entropy to the focus loss function,and increase the detection ability of the parts by setting the scale weight to the samples that are easy to be misclassified.The final recognition m AP reached 92.49%,and FPS was 6.75,which reached the detection accuracy required by the system.(2)Aiming at the problem of sample imbalance in abnormal detection of catenary parts-normal samples are far more than abnormal samples,a zero sample based abnormal detection strategy based on Ganomaly is proposed.Use the normal samples in the catenary parts to train the network,and let the network learn that it normally performs automatic coding conversion on the images input by the network.If the distance between the conversion potential and the coding potential variable is less than the threshold,the image is considered to be abnormal,otherwise It is believed that there are abnormalities in the parts in the picture.In this paper,the network structure has been modified and layer jump connections have been added,so that the effect of the network model has been improved to a certain extent.(3)Another strategy for the imbalance between positive and negative samples,using the DCGAN network to enhance the data of each part detected by the contact network,resulting in the number of normal samples and abnormal samples tending to balance,and then use the data-enhanced pictures Training classification network,the accuracy of Res Net classification network reaches 91% respectively.Here,select the improved YOLO detection network and ResNet classification network as the core algorithm of the anomaly detection system.Based on the above research content,this paper designs and implements the highspeed rail contact network anomaly detection system.Using Python to develop a crossplatform system,to ensure that the system can run on multiple platforms.By using a large number of real pictures to test the system,the results show that the detection system can run stably and achieve the expected detection accuracy.It proves that the high-speed rail catenary anomaly detection system has a reasonable design can accurately and effectively detect abnormal parts.