Association Rules Mining And Analysis For High Speed Railway Overhead Contact System Fault Data

In recent years,China’s high-speed railway has developed rapidly.With its annual mileage and passenger traffic volume continuously increasing,it has now ranked first in the world and makes an important contribution to the development of national economy and daily convenient travel.As an important part of high-speed railway traction power supply system,overhead contact system(OCS)is the power source of the train,and its good operation state plays an important role in ensuring the safe and reliable operation of high-speed railway.At present,with the increase of OCS service time and the improvement of detection equipment,a large number of fault records are collected and accumulated in the operation and maintenance process.These data can truly reflect the characteristics of the OCS.How to effectively mine interesting information has become an urgent problem.Reliability assessment,as the main method to study the fault data of OCS,has made great contribution to evaluating OCS.However,due to the variety of OCS faults,the fault tree cannot contain all fault items,otherwise it will be too complex.However,association analysis,one of big data technology,has no limit on the number of fault items,which provides the possibility to study the historical fault data from the system-wise perspective.Therefore,this paper uses association analysis technology and deeply studies the process of existing algorithms.Two algorithms are proposed to realize the mining of historical fault data according to the characteristics of high-speed railway OCS fault data.Besides,reference suggestions are provided for maintenance decision-making according to the mining results.The main work of this paper is as follows:Firstly,this paper studies the data structure and coding theory,understands the fault items and mechanical composition of OCS.By selecting the appropriate data structure,a five-level tree structure is constructed for OCS fault items,and a complete and scalable coding system is established based on this tree structure.Data preprocessing and encoding data are carried out on the fault records which are collected from several railway administrations.Secondly,this paper studies the existing algorithms of association analysis and proposes the multi-dimensional information partition model according to the sparsity of OCS fault data,which can flexibly transform fault records to transaction table and reduce the sparsity of data.The fault intensity-based hierarchical index association analysis model(FIHI)is designed based on the Apriori framework according to the hierarchy of OCS fault data,which can effectively obtain the frequent itemsets and association rules from OCS fault data.Thirdly,this paper studies the redundancy of frequent itemsets in mining results and proposes a brand-new concept,marginal frequent itemset.Based on this concept,a new algorithm is designed which changes the task of mining frequent itemsets to mining marginal frequent itemsets with the purpose of simplifying the mining results and getting more accurate location of fault.The results of experiment show that this new algorithm is not only better than the original algorithm in mining results,but also better in algorithm efficiency.Finally,this paper studies the generated association rules in mining results and proposes multiple filtering mechanisms to realize the effective filtering of rules.The association knowledge base is established by the filtered association rules and the updating method based on time slicing is proposed to maintain the timeliness of the associated knowledge base.According to the mining results,the fault relation network is constructed,and two indexes,fault-in-degree and fault-out-degree,are used to analyze the impact of fault items on the network quantitatively,so as to provide suggestions for maintenance decision-making.