Data Mining Based Alarm Correlation Analysis In EMU

As an important part of high-speed railway,the number of EMUs in China continues to increase,and the level of EMU integration and intelligence is significantly improved,which puts forward higher requirements for EMU maintenance.At present,different types of EMU are equipped with fault alarm devices and recording devices,which are used to find and record the faults that occur during the operation,so as to find the problems in maintenance in time.However,the phenomenon of “domino” exists in EMU fault alarm,and the traditional manual troubleshooting cannot meet the development requirements of intelligent maintenance.Alarm correlation analysis aims to obtain the correlation knowledge between alarms from large amount of alarm data,by using these alarm correlation knowledge,the propose of narrowing the fault scope and accurately locating the root cause of the fault can be achieved.At present,the big data characteristics of EMU fault alarm appear.Alarm correlation analysis based on data mining can replace manual discovery of fault alarm correlation knowledge,reduce the dependence of knowledge acquisition on expert experience,and introduce the Hadoop platform into alarm correlation analysis,which can improve the efficiency of rule mining.In this paper,according to the characteristics of EMU fault alarm data,the research scheme of EMU fault alarm correlation is developed.Data mining technology and Hadoop platform are used to mine fault alarm correlation knowledge,which provides knowledge support for EMU intelligent maintenance.The main research work of this paper is as follows:(1)This paper organizes and deeply analyzes the fault alarm data of EMU,summarizes the data characteristics and their existing correlations,furthermore,combine the EMU fault alarm data characteristics to formulate a fault alarm correlation research program suitable for EMU.(2)In view of the problems of data redundancy and time asynchrony in alarm data,a dynamic sliding window method based on density clustering is used to transform the data,in addition,study the scheme of parameter selection,and verify the efficiency of optimization algorithm and the accuracy of parameter selection.(3)Aiming at the problem of unequal importance of EMU fault warning items,the weighting idea is introduced into the FP-Growth algorithm.To solve the problem of low efficiency of FP-Growth algorithm in mining skewed data,the weighted FP-Growth algorithm based on Map Reduce is studied and implemented.(4)In order to eliminate the subjective factors of artificially setting alarm item weights and artificially setting weighted support thresholds,which adversely affects the results,the entropy method is used to calculate the weight of the alarm items by considering multiple attributes of the alarm data,at the same time,the binomial distribution method is introduced to calculate the count expectation of the alarm items,and weighted support threshold is set accordingly.(5)A Hadoop platform is built to verify the performance of the algorithm.The experimental results show that the weighted FP-Growth algorithm based on Map Reduce has high mining efficiency.When the database is expanded,the algorithm also has good scale growth and spatial scalability.Finally,the alarm correlation analysis is carried out on the fault alarm data of a certain vehicle type of EMU,and the correlation knowledge of EMU fault alarm is obtained.The fault range is reduced and the root fault is accurately located by using the correlation knowledge of fault alarm.