Research On Fault Combination Prediction Method For On-Board Equipment Of CTCS Based On Cross Entropy Theory

As an important part of high-speed railway train operation control system(hereinafter referred to as train control system),on-board equipment is the core component to ensure the safe and efficient operation of the train.Among them,300T on-board equipment of the train control system occupies a high proportion in the on-board equipment currently online.Due to its characteristics of complex structure,numerous modules,long time high-speed operation across regions and variable driving environment,any minor fault or hidden danger may reduce the reliability of train operation.Therefore,it is necessary to timely predict the fault state of the on-board equipment in the train control system,identify and eliminate potential faults in advance,and realize condition based maintenance,so as to ensure the safety and operational efficiency of high-speed trains.At present,the research for on-board equipment failure of the train control system mainly concentrated on the fault type analysis,fault diagnosis and identify.There is a lack of research on related theories,methods and technologies for fault prediction of on-board equipment of train control system,especially there is no practical fault prediction theory and method which can be applied in the field.The field fault data processing intensity is extremely great.In order to relieve the pressure of fault analysis and treatment on site and realize condition based maintenance,it is very necessary to carry out the research on fault prediction of on-board equipment of train control system.This thesis bases on the log data of 300T on-board equipment of train control system.Firstly,the log data of on-board equipment is cleaned and the multi-fault mode for on-board equipment is analyzed.Then,the failure statistical characteristics for component-level and system-level failure prediction is built.The effective identification of fault running state is realized.A fault combination prediction method for the train control system of on-board equipment based on cross entropy theory is designed.The main work of the thesis includes:(1)The characteristics of 300T on-board equipment log data is analyzed,and the data preprocessing strategy is designed to realize fault data statistics accurately.Multi-state system theory is used to divide the running state of on-board equipment.Based on the fuzzy FMEA method,the running state evaluation of on-board equipment and the accurate classification of fault types within fault levels are realized.(2)Failure rate and equivalent failure rate are selected as the fault characteristics of on-board equipment at component level and system level respectively,and support vector machine,extreme learning machine and grid search optimization support vector machine are used to identify the running state of on-board equipment.Simulation results show that the equivalent failure rate can effectively represent the running state of on-board equipment,and the grid search optimization support vector machine has the highest state recognition accuracy of 93%among the three methods,which verifies the feasibility of using this method for on-board equipment state recognition.(3)The equivalent failure rate and failure rate data series are used as data support.Firstly,the k-nearest neighbor non-parametric regression prediction model and the improved grey Elman neural network prediction model are used for short-term prediction,respectively.Secondly,based on the cross entropy theory,the two models are optimized and combined,and the common equal-weight combination model is selected for comparative analysis to predict the fault time at the component level.Finally,the predicted value of equivalent failure rate is identified by grid search optimization support vector machine,and then the fault time,fault level and fault module at the system level are predicted.Simulation results show that compared with the other three kinds of prediction model,the prediction results of cross-entropy combination prediction model has the highest prediction accuracy,the prediction accuracy of fault grade is 87.33%,the prediction accuracy of fault module is 94.4%,the prediction accuracy of fault mode is 77.77%.The feasibility of using cross-entropy combination prediction model for fault prediction of on-board equipment is verified.In this thesis,on the basis of systematically analyzing the log data of on-board equipment,a fault combination prediction method based on cross entropy theory is introduced to predict fault of on-board equipment,which can effectively realize the prediction of failure time,failure grade,fault module and fault mode of on-board equipment.The prediction accuracy basically meets the demand of fault analysis and prediction in the data processing working area of the high-speed train section,which lays a foundation for condition based maintenance of on-board equipment.This thesis contains 58 figures,37 tables and 82 references.