Research On Detection Method Of Train Holding Brake Based On Multi-source Data Fusion

Railway freight is an important part of China's logistics,undertaking a large number of transportation tasks,and the safety of railway freight trains has always been attached great importance.The brake failure of railway freight train is a kind of safety problem of "the highest frequency,the biggest security threat" in current railway transportation.The brake detection is the most important train safety detection project of railway department,so it is of great significance to study the brake detection method of railway freight train.The existing train holding brake detection methods mainly include manual detection and single equipment detection.The data source is single,which can not accurately identify the phenomenon of train holding brake,and the rate of missing report and false report is high.By collecting and integrating relevant information,this paper analyzes the principle of freight train braking and the phenomenon of train holding brake,studies the application status of existing detection methods of train holding brake,finds out the shortcomings of existing detection methods and clarifies the necessity of this paper;Through the research on the theory and application of multi-source data fusion,the significance and feasibility of train holding brake detection method based on multi-source data fusion are confirmed.Therefore,this paper proposes a train brake detection method based on multi-source data fusion.This method integrates the detection results of several single equipment,such as hot wheel detector,abnormal sound detection system and wheel foreign body detection system,and integrates the data of speed sensor to make full use of the multi-source data of each single detection equipment for train brake detection.Methods input the test results and sensor data of each single train holding brake detection system,and output whether the train is holding brake or not.The specific implementation steps are as follows: firstly,the sample database is established,the relevant features in the sample database are extracted,and the features are fused to form a fused sample data set;Then,the train holding brake detection model based on isolation forest anomaly detection algorithm,one class SVM single classification support vector machine anomaly detection algorithm,and RUS-SVM classification algorithm combined with random Under Sampling and support vector machine is constructed.The missing and false alarm rates of the model on the fusion data set are calculated and compared with the existing train holding brake detection methods.The empirical results show that,compared with the existing train holding brake detection methods,the proposed train holding brake detection method based on multi-source data has lower false alarm rate and false alarm rate.The false alarm rate and false alarm rate of the proposed train holding brake detection model based on RUS-SVM classification algorithm are less than 6%,which has good promotion value.