Abstract: To mining process models from different periods of business process change and improving accuracy of mining results, a new method of adaptive process mining based on time-varying sliding window was proposed. The related concepts such as time-varying sliding window and adjacent event probability dependency were defined on the basis of analyzing characteristics of business process change. The update rules of modifying size and progress in a time-varying sliding window through controlling process instance time were studied. A new process mining algorithm as well as process model mining rule were presented based on adjacent event probability dependency, and process instances in a time-varying sliding window were modified continuously according to changing frequency and arrival rate from mined results and process instance streams, thereby process models from different periods of business process change were mined. The experiment results indicated that the adaptability and anti-noise property of proposed method were better than existing process mining method.

Key words: process mining, adaptability, adjacent event probability dependency, multi-period, time-varying sliding window

摘要： 为了挖掘流程变迁过程中各时段的流程模型，提高流程挖掘结果的准确性，提出了一种基于时变滑窗的自适应流程挖掘方法。在分析了业务流程变化特点的基础上，定义了时变滑窗、相邻事件概率依赖关系等相关概念，研究了以流程实例时间为变量,来控制调整滑窗大小和滑动进度的日志更新规则；基于相邻事件概率依赖关系，给出了流程模型挖掘规则和一种新的流程挖掘算法，根据流程挖掘结果的变化频度和流程实例日志流的到达速率推动时变滑窗持续变更，进而挖掘出整个流程日志中各时段的流程模型。实验结果表明，与已有的流程挖掘方法相比，所提方法具有良好的自适应性和抗噪性。

关键词: 流程挖掘, 自适应性, 相邻事件概率依赖, 多时段, 时变滑窗