System Log Conformance Checking Based On Process Mining

Process mining has become a method to analyze organizational behavior.It extracts knowledge from event logs and provides technology to discover,monitor and enhance actual processes.In the past decade,it has been used in business process management(BPM)and data mining.With more and more application of process aware information system,a large number of event data are recorded,which can be analyzed by process mining technology.Generally,enterprises do not model systematically in the initial stage,but a large number of activities are still recorded in their systems.With the gradual complexity of business,enterprises usually use some existing process mining technology to mine workflow models in a large number of actual logs to better manage business,so as to improve work efficiency,but we often need compliance detection to measure the quality of the mined models.Organizations are likely to have similarities,differences and diversities between their business processes.In order to coordinate and make these processes effective and efficient,it is necessary to analyze the similarities and differences between existing processes.Business process similarity(BPS)is an activity to identify the similarity between two different business processes.At the same time,the implementation of process aware information system supports process model configuration during construction and process instance change at runtime.Their adaptation leads to a large number of process model variants,which are derived from a process model,but slightly different in structure.In general,it is very expensive to configure and maintain such model variants.The research on the similarity between process model and its process variants is beneficial to reduce the requirements of process variants for future process configuration and adaptation,so as to facilitate maintenance and reduce costs.Based on the theory of Petri nets,this paper analyzes the similarity of process model and its process variants,the similarity between different business processes,combining the relationship between behavior profile knowledge and activity,and gives the measurement method of the conformity between event log and process model.(1)Aiming at the problem that the process variants with the same distance can not be further distinguished in the existing research,a distance similarity analysis method based on behavior contour is proposed by using the powerful analysis function of behavior contour in describing process model similarity.On the basis of equal model distance of process family,the similarity degree of behavior contour is introduced,and the structural and behavioral characteristics of process family model are comprehensively evaluated by using two-dimensional measurement indicators.The similarity degree of process models and their process variants are more comprehensively analyzed,so in order to solve the problem that the existing literature cannot distinguish the differences among process variants by using a single dimension.(2)In the process mining domain,event data is used to discover process models,and conformance detection can evaluate the quality of mining models.In view of the fact that most of the existing conformance detection methods are based on the alignment of single activity,and only the fast matching method between the model and the log is given,and the non-block structure model is not sensitive,a process conformance detection method based on the alignment of direct successor relationship is proposed.First,according to the direct follow-up relationship,the related attributes in the model and the log are displayed in the form of the nearest activity pair.Then,according to the optimal alignment algorithm AAP based on the nearest activity pair,the optimal matching activity pair sequence of the trajectory is obtained,and the fitness function of the single trace and the model based on the minimum cost is proposed,and the fitness algorithm DFA is given to measure the consistency between the log and the model.(3)In view of the existing process model behavior similarity measurement methods are mostly based on the occurrence sequence of changes,which has the problem of high time complexity,this paper proposes a process similarity measurement method based on the occurrence relationship of activities.First,the concept of left and right sets is proposed,and the violation degree and the weight of left and right sets are introduced.According to whether the relationship between activities changes,the weight value is determined and the similarity of activities is calculated.Then,the similarity of activities in the process is normalized,the definition of process similarity is given,and the process similarity algorithm(AOR)based on the relationship between activities is proposed to carry out the phase of business process Similarity measure.There are two main experiments in this paper,which are used to verify the fitting algorithm based on direct successor relationship alignment and the process similarity algorithm based on activity occurrence relationship.The two experimental results show the feasibility of the two algorithms.In addition,the performance of the fitting algorithm and the similarity algorithm is tested by the model compiled manually,and the running time of the algorithm is recorded.Compared with the existing mainstream algorithm,the algorithm in this paper is better than other algorithms to some extent.The main research contents and contributions of this paper are as follows:(1)A similarity measurement method of process model and its variants is proposed to effeectively distinguish the process family model with equal distance.(2)A process similarity algorithm based on the relationship between activities is designed.This method combines the structural characteristics of activities on the left and right sets with the changes of the relationship between activities,introduces the weight of the left and right sets,and comprehensively analyzes the structure and behavior to measure the similarity of different processes.(3)Based on the direct follow-up relationship,the concept of the alignment of the nearest active pair is proposed for the first time,and the optimal alignment algorithm based on the nearest active pair is proposed to find the optimal alignment of the trace to be matched and the model execution trace.The cost function based on the optimal alignment is defined,and then the fitting degree between the model and the chronicle is calculated.This method can quickly match the execution trace and the trace to be matched,and gives a method to measure the consistency between the model and the log.Figure[12]table[9]reference[79].