Research On Deviation Detection And Model Repair Based On Event Log

With the advent of the era of big data,the event logs generated by various information systems become more and more complex.It is very important to mine useful information from these event logs.Process mining techniques aim to build process models from event logs to manage,monitor,and enhance actual business processes.However,due to various factors such as changes in external environment and requirements,business processes will be constantly updated,which makes the process model initially constructed unable to accurately reflect the actual operation of business processes.If the process model is re-mined from a large number of event logs,it will not only produce a very high cost,but also cannot guarantee that the mined model is highly similar to the existing model,which makes the existing model lose its value.Therefore,model repair technology comes into being.How to accurately locate the deviation in the existing model and properly repair the model on the premise of preserving the behavior relationship of the existing model as much as possible has become the key problem of model repair.Most of the existing bias detection and model repair technologies only consider the control flow,ignoring the information contained in the activities and the logical relationship between the activities.As a result,the repaired model is either very complex or has low accuracy.In this context,this paper carried out research on the event log deviation detection and model repair methods based on infrequent behaviors and behavior profiles,and improved and innovated the existing deviation detection and model repair methods.The details are as follows:(1)A business process modular modeling method based on behavior profile is proposed.For the process model containing multiple systems,the use of complete event log modeling may occupy a lot of resources and is prone to deviation.In this paper,a complete business process system is first decomposed into different operating modules,and then modeled according to the event log in each module and the behavior profile relationship between activities.The process model based on Petri net is constructed by PM4 Py,and then the modules are combined into a complete process model by using the behavior relationship between modules.Finally,the model is analyzed and optimized.The feasibility of the modular modeling method is verified by the simulation experiment.The experimental results show that the optimized model is more reasonable and efficient than the original model.(2)The research of deviation detection and model repair method based on infrequent behavior is proposed.In view of the problem that the current repair method directly deletes the infrequent behaviors in the event log as "noise",which leads to the loss of the important information contained in these infrequent behaviors in the process model,this paper firstly adopts the workflow into-output network(WFIO network),which integrates data flow and control flow,for modeling,and analyzes the characteristics of the infrequent behaviors from the perspective of data flow.Three deviation detection methods are proposed.Then,different types of deviations were located by the improved actively-data association matrix(ADIM),and different model repair strategies were proposed for different types of deviations.The experimental results show that this method can effectively detect the deviation caused by infrequent behavior,and the repaired model can better reflect the actual business process.(3)A modular repair method based on causal behavior profile is proposed.In view of the fact that the existing model repair methods cannot properly weigh the fitness and accuracy of the model,and the cost of repair is high,this paper divides the model into several model segments through decomposition,and then proposes the causal behavior contour relationship between activities in the event log.By comparing with the causal behavior contour relationship between activities in the model,Finally,a model fragment repair method based on the causal behavior contour relationship in event logs is carried out to repair one or more model fragments with deviation,so as to realize the whole model repair.The experimental results show that this method can improve the accuracy of the repaired model as much as possible under the condition of high fitness.Figure [31] Table [20] Reference [90]...