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${}_{\mathrm{CL}}$, is developed for mining cross-level marginal frequent itemsets from railway OCS fault database. Our performance study shows that MFIM${}_{\mathrm{CL}}$ 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.

架空接触系统（OCS）作为电气化铁路的动力源，其组成复杂，故障种类繁多，对其故障预防提出了很高的要求。近年来，随着铁路OCS故障数据库的建立，关联分析被用于从系统角度实施故障预防，为运维提供指导。然而，由于故障数据库的层次结构，现有的频繁项集挖掘在结果中存在大量冗余，无法定位最精确的故障，影响决策，使得故障排除缺乏针对性。为了解决这个问题，本文提出了一个新概念，称为边际频繁项集，这是一个由满足阈值的分层数据库中尽可能精确的项目组成的项目集，以及另一种挖掘任务：挖掘边缘频繁项集而不是所有频繁项集。提出了路径变换和项集的递减深度两种方法来实现边际频繁项集的挖掘。提出了边际度和边际信息量两种新的度量方法来评价挖掘结果的内容。一种高效的算法，命名为 MFIM 建议对挖掘结果的内容进行评估。一种高效的算法，命名为 MFIM 建议对挖掘结果的内容进行评估。一种高效的算法，命名为 MFIM${}_{\mathrm{CL}}$, 是为从铁路 OCS 故障数据库中挖掘跨级边际频繁项集而开发的。我们的绩效研究表明，MFIM${}_{\mathrm{CL}}$具有较高的性能，可以获得更多的关键信息并减少结果的数量。此外，边际频繁项集挖掘可以简化关联规则构建的故障关系网络，优化铁路OCS故障预防的决策过程。