动车组故障诊断知识挖掘中改进的并行频繁模式增长算法

Improved parallel frequent pattern growth algorithm in EMU's fault diagnosis knowledge mining

针对动车组历史运维数据的知识挖掘问题,从有效利用动车组历史运维数据来指导动车组故障诊断的角度出发,分析了现有并行频繁模式增长算法的实现形式和不足。结合动车组故障诊断的要求,提出利用局部频繁模式树代替全局频繁模式树的数据挖掘算法。该算法在各主要步骤上均实现了并行处理,优化了局部频繁模式树生成规则,对频繁模式的搜索策略进行了改进。改进后的算法大大提高了关联规则挖掘的效率,挖掘结果很好地保留了故障信息与状态信息之间的关联关系,并合理去除了无效规则。通过对该算法的具体分析与实际测试,表明该算法在动车组故障诊断知识获取过程中具有快速、高效、准确的特点。

Aiming at the knowledge mining problem of Electric Multiple Units （EMU） trains＇ historical operation and maintenance data, the existing parallel frequent pattern growth algorithm and its disadvantages were analyzed from the perspective of guiding EMU＇s fault diagnosis. According to the requirements of EMU＇s fault diagnosis, an im- proved algorithm for data mining which used local frequent pattern tree instead of global frequent pattern tree was proposed. In this algorithm, the parallel processing in every data processing steps was adopted, the production rules of local frequent pattern tree were optimized, and the search strategy of frequent patterns were also improved. The efficiency in the process of mining association rules was greatly improved by the proposed algorithm, the relationship between fault information and state information was kept well by the mining results, and the invalid rules was re- moved reasonably. Based on analysis and experimental tests, the characteristics of fast speed, high efficiency and ac- curacy of the proposed algorithm in the process of knowledge acquisition of EMU＇s fault diagnosis was illustrated.