Abstract: Existing process discovery approaches construct a flat process model by mining directly-follow relation,concurrency relation,loop relation,and choice relation between activities.However,these approaches have difficulties in handing multi-instance sub-processes.To overcome this problem,a Hierarchical Multi-instance Processes Mining (HPM2) approach was proposed to support the discovery of business processes with multiple sub-processes instantiations.The nested relation between activities from the event log with multi-instance sub-processes was mined,and the hierarchical event log was constructed.Then the identification and reconsitution of sub-processes multi-instance on sub-log was implemented,the existing process discovery approach was used to mine the sub-processes model,and finally the model of hierarchical multi-instance Petri nets was found.The proposed approach had been implemented in the open-source process mining toolkit ProM.Based on public datasets,the HPM2 approach was compared with the state-of-the-art process discovery approaches systematically,and the advantages of the proposed approach in deal with event logs with multi-instance sub-processes information was further verified.

Key words: process discovery, hierarchical business processes, multi-instance identification, Petri nets, quality evaluation

摘要： 针对已有过程挖掘方法通过挖掘任务之间的顺序、并发、循环和选择关系构造的扁平过程模型,很难处理带有多实例子过程信息的事件日志,提出一种分层多实例过程模型挖掘方法HPM2。首先从带有多实例子过程信息的事件日志中挖掘任务间的嵌套关系并构造分层事件日志;然后对子过程的事件日志进行多实例识别与重构,应用已有模型挖掘方法进行子模型挖掘,最终发现分层多实例Petri网模型。所提方法均已在开源过程挖掘平台ProM工具中实现。基于公开事件日志数据,系统比较了HPM2方法与已有过程挖掘方法挖掘模型的质量,进一步验证了所提方法处理带有多实例子过程信息事件日志的优势。

关键词: 过程发现, 分层业务过程, 多实例识别, Petri网, 质量评估