Identification And Analysis Of Deviation Domain Of Business Process Model Based On Log Filtering

The rapid development of big data technology has had a far-reaching impact on the management and optimization of business processes in enterprises.As an important research topic of enterprise business management,process mining needs to extract information from the event log and establish a real execution model.However,many process mining technologies work under the assumption that the behaviors related to the underlying process execution are correctly stored in the event log.In fact,there are some exceptions in the real event log,including recording errors,human errors,imprecision of recording tools and other factors.These factors lead to the inaccuracy of the mined business process.Therefore,filtering noise before mining process is of great significance.At the same time,in the process of actually mining the process model,there will be some deviation domains and errors between the event log and the mined model.Therefore,preprocessing the log and analyzing the deviation domains between the process model and its corresponding execution log to realize the effectiveness and correctness of mining the business process has become a hot topic of business process management.At present,the research on low-frequency noise filtering mainly depends on the frequency dependence and behavioral profile relationship of event log,and few consider low-frequency behavioral from the perspective of frequency change of different sequence relationships.In order to make up for the limitations of existing methods,the event log is filtered based on frequency change rules and string matching algorithm under the premise of incomplete log.At present,the research on conformance mainly depends on the technology of replay and alignment,which are proposed based on the alignment of a single activity,while the research based on the alignment between activity pairs is relatively few.The existing technology has some disadvantages for the model with complex structure.Therefore,this paper proposes a conformance checking method based on the deviation domain of activity pair pattern recognition.Using the activity pair pattern to identify the local change deviation domain between the original event log and the actual sequence of the model,this method can deal with the complex process model with loop structure and concurrent structure.On the basis of filtering log noise,this paper mainly analyzes the deviation domain between the mined process model and event log and modifies the model.The main research is as follows:(1)The business process model is optimized based on the the transition occurrence rules and the behavioral profile basic relationship of Petri net.Using the case of a fever clinic,the optimized process model reduces the complexity of the model and avoids cross-infection,and maximizes the utilization of medical materials.Finally,the PIPE software is used to analyze the state space and reachability diagram of the process model to verify that the optimized model is bounded,safe,deadlock-free and improve efficiency.(2)Aiming at the problem that low-frequency logs are directly regarded as noise filtering in process mining,a business process low-frequency log noise filtering method based on string matching algorithm KMP is proposed.On the basis of the directly-follows graph and the eventually-follows graph theory,the abnormal structural fragments in the directly-follows graph of the low-frequency log are identified according to the frequency change rules,which correspond to the sequence set of invalid direct activity pairs in the event log.Combined with the string matching algorithm,the invalid activity sequence is matched with the low-frequency log trace,and the corresponding log trace is directly filtered after the matching is successful,so as to filter the noise in the low-frequency log.(3)Aiming at the deviation between the process model and the observation behavior,a conformance checking algorithm based on activity pair pattern recognition deviation domain is proposed.On the basis of the activity pair model theory,the process model structure is divided into two types: non-concurrent structure and concurrent structure by using the behavioral profile of the process model,Combined with the complete prefix expansion technology,the pattern sets of different specified activity pairs are calculated in the finite Petri net.Then,the local change deviation domain between the original event log and the actual sequence of the model is identified,and the model is modified.Figure [28 ] Table [11] Reference [80]...