Deviation Detection And Repair Techniques Of Process Models Based On Event Logs And Logic Petri Nets

With the development of big data technology,information systems have become an important tool to manage business processes.Process mining technologies can extract knowledge from the event logs generated by information systems for process discovery,conformance checking and process enhancement.Process discovery can build process models based on the information contained in event logs.Conformance checking is used to analyze whether there are deviations between process models and event logs.Process enhancement uses event logs to extend or improve the existing models.As an application of process enhancement,model repair technologies can repair process models according to event logs,making the repaired models reflect the actual business processes correctly.As the business processes changing,the application of model repair techniques become more and more extensive.The existing model repair methods focus on adding self-loops and sub-processes,which make the repaired models more complex and redundant.Therefore,it is an important research to precisely find deviation positions and improve the qualities of repaired models.For different types of event logs,this paper proposes different deviation detection and model repair techniques based on logic Petri nets and order relations among activities.For the event log containing new activities,if there are parallel or sequential relations between new activities and original ones,a ladder matrix and process tree-based deviation detection and model repair method is proposed.It adds new activities as branches to parallel structures,or constructs new parallel structures or sequential ones in the model.First,we find the devaiations between an event log and a process model by ladder matrix,and generate a deviation matrix.Then the deviation matrix and process tree are used to add new activities to the model,and the repaired model can be obtained.For the event log containing new activities,if there are choice relations between new activities and original ones,another deviation detection and model repair method based on deviation set is proposed.It aims to add new activities as branches to choice structures,or construct new choice structures in the model.We use log order set and model order set to record the order relations in an event log and a process model,respectively.A deviation set is generated by comparing the two sets.Then,we add new activies to the model based on the deviation set,and add logic functions according to the order relations among activities,and we can obtain a logic Petri net-based repaired model.For the event log not containing new activities,if there are parallel relations among activities in the event log,while there are choice or sequential relations among transitions in the model,this paper designs a deviation detection and model repair method based on comparison set.It replaces choice or squential structures with parallel structures.First,the problematic structures are found by comparison set.Then we change improper structures to correct structures,and add some logic expressions to describe the relations among activities.Finally,a repaired model can be got.For the event log not containing new activities,if some activities belong to different choice branches in the event log,while the model only replays activities in the same choice branch,a deviation detection and model repair method based on deviation sub-log and process tree is proposed.It aims to jump from a branch to another branch in a choice structure.Firstly,we identify choice structures by process tree.Then a choice devistion sub-log is obtained from a given event log.According to this sub-log,we add direct arcs among different branches in a choice structure,and add logic expressions to control the jump behaviors between different choice branches.Thus,we obtain a logic Petri net-based repaired model.Finally,we illustrate the effectiveness and correctness of the proposed methods by simulation experiments and comparative analysis.The repaired models by our methods have ideal simplicities and precisions while ensuring high fitnesses.