Marginal frequent itemset mining for fault prevention of railway overhead contact system.

The overhead contact system (OCS), as the power source of electrified railway, has a complex composition and various types of faults, so it places high requirements on its fault prevention. In recent years, with the establishment of railway OCS fault database, association analysis has been used to implement fault prevention from system-wise perspective and provide guidance for operation and maintenance. However, due to the hierarchical structure of fault database, the existing frequent itemset mining has a lot of redundancy in the results, and cannot locate the most precise faults, which affects the decision-making and makes troubleshooting lack of pertinence. To address this issue, this paper proposed a new concept, called marginal frequent itemset, which is an itemset composed of as precise items as possible in hierarchical database that meets the threshold, and an alternative mining task: mining marginal frequent itemsets instead of all the frequent itemsets. Two methods, path transform and descending depth of itemset, are proposed for achieving mining a set of marginal frequent itemsets. Two novel measures, margin degree and marginal information quantity, are proposed to evaluate the content of the mining results. An efficient algorithm, named MFIM _CL , is developed for mining cross-level marginal frequent itemsets from railway OCS fault database. Our performance study shows that MFIM _CL has high performance and can obtain more key information and reduce the number of results. Furthermore, marginal frequent itemset mining can simplify the fault relation network constructed by association rules and optimize the decision-making process for fault prevention of railway OCS.
Competing Interests: Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
(Copyright © 2021 ISA. Published by Elsevier Ltd. All rights reserved.)