Low-frequency Behavior Mining And Optimization Analysis Of Business Process Based On Petri Net

With the explosive growth of process log data,the analysis and research of business processes is becoming more and more common.From simple synthetic event logs to complex large-scale real event logs,algorithms for data mining are becoming more and more complex.For such problems,process mining technology can closely link event data and process models to better improve and facilitate the development of business processes.As the main content of business process analysis,process mining mostly finds frequent behaviors in research,and rarely involves low-frequency behavior,but low-frequency behavior is also important for process management.In general,process mining is designed to mine business process models by discovering process models that fully reflect the behaviors seen in the event log.However,the low-frequency behavior is inevitably included in the business process model.If the low-frequency behavior is directly ignored during the process of mining low-frequency behavior,the business process model will lose some of the rules and affect the structure of the business process model.Therefore,for the problem of low-frequency behavior in business process,this paper mainly introduces the mining of fusion module network and feature network,the analysis method of effective low-frequency pattern based on behavioral closeness,the method of optimization and analysis of low-frequency behavior based on process tree cutting,optimization method of low frequency behavior based on integer linear programming and so on.In the process of mining low-frequency behavior,it is necessary to consider not only the mining of simple event logs,but also the mining of interaction process models between different modules;not only the low-frequency behavior from data attributes,but also the behavioral attributes of communication characteristics between different modules to analyze low frequency behavior.The main contributions of this paper include:(1)Aiming at the mining of interaction process models between different modules in process mining,a method based on fusion feature network and module network mining low-frequency behavior patterns is proposed.This method overcomes the difficulty of mining low-frequency behavior of interactive process models in a simple event log and is able to find all low-frequency behavior patterns.By processing the effective event log to determine the communication behavior profile relationship,according to the log,the features are divided into different modules,the intermal behavior of the event is reconstructed,and the corresponding module network and feature network are mined;the feature network and the module network are merged to obtain a complete process model.And all the low frequency patterns are derived by iteratively expanding the initial pattern.(2)Aiming at the problem of low-frequency behavior directly filtering as noise in process mining,an effective low-frequency pattern analysis method based on Petri net behavior closeness is proposed.Based on the Petri net behavioral profile theory,the pre-event log is used to establish the initial process model.The frequency of the pattern is calculated according to the process model and all low-frequency behavior patterns are found compared with the threshold.Calculate the behavioral closeness between the log and the model in the low frequency pattern to mine an effective low frequency behavior pattern.(3)Aiming at the mining problem of low-frequency behavior bet,Areen different modules in process mining,a method based on Petri net for mining and optimizing low-frequency behavior of business process is proposed.Most of the existing mining methods are to study low-frequency behavior from data attributes,and to rarely analyze low-frequency behavior based on behavioral attributes between different modules.For this problem5 based on the communication behavior profile between modules,this chapter mines the initial process model by preprocessing the event log,and matches the initial model with the direct flow graph of the cut-off relationship to find the low-frequency behavior.The effective low-frequency log and noise log are distinguished based on the behavioral closeness of the log and the model,the noise is filtered,and the event log is optimized.By using an event log that does not contain noise sequences,a sound optimized Petri net model is obtained by merging the interactive module network and the feature network.(4)Aiming at the problem that the low-frequency behavior can not be handled well by using integer linear programming,a domain-based method for mining and analyzing low-frequency behavior of business processes is proposed.This approach breaks the limitations of using the simple event log as input in the process discovery process.The pre-processed event log is used to find the corresponding prefix closure,and using the prefix closure language to construct the corresponding domain and linear inequality,find a reasonable process model based on the obtained solution,and use condition-dependent metric relationship to distinguish low-frequency behavior and noise,and then filter noise to obtain an optimized event log without noise.Figure[19]Table[9]Reference[102]...