Abstract: In model checking and process mining,trace clustering groups similar execution traces to support the construction of accurate behavioral models,the verification of model correctness,and data-driven model refinement.However,existing sequence pattern-based trace clustering approaches typically treat traces as ordinary strings,neglecting the inherent concurrent and cyclic execution relationships among activities.This simplification often leads to the loss of structural information and consequently degrades clustering quality.To address this issue,this paper proposes a novel trace similarity measurement method,referred to as process edit distance.The proposed method first normalizes concurrent execution relationships among activities into a consistent sequential representation.It then abstracts repetitive loop behaviors through a compression and simplification mechanism to reduce the influence of redundant executions.Finally,trace similarity is measured by jointly considering activity occurrences and the direct-follow relationships between activities.Furthermore,to obtain process models that better reflect real business behavior,a post-processing strategy termed merging noise clusters is introduced within an agglomerative hierarchical clustering framework to alleviate structural fragmentation caused by noise or small-sized clusters.Experimental results demonstrate that the trace clustering algorithm based on process edit distance outperforms existing methods of the same category in terms of clustering quality,while exhibiting strong stability and robustness.In addition,the merging noise clusters strategy consistently reduces the overall structural complexity of the clustering results,leading to clearer and more interpretable process models.

Key words: Model checking, Sequence patterns, Trace clustering, Process edit distance, Merge noise clusters