| Process mining is a kind of tracking and analysis of process technology,with the development of the process of mining technology,more and more large enterprises begin to apply the process mining techniques from the system automatically records the event log mining knowledge to detect,monitor and improve the actual process,which not only reduced the management cost,also greatly improve the efficiency of enterprise operation and management.In practical applications,the system will generate a lot of flow logs because the flow runs continuously.Manual observation makes it difficult to detect a large number of logs,especially to detect anomalies.It requires a way to detect anomalies in large logs that have a significant impact on the results of process mining.How to deal with abnormal behavior in event logs is a key issue in process mining,which is related to the quality of mined models.However,many previous studies on abnormal behavior of event logs only focus on the control flow structure,focusing on the sequence and frequency of activities,and ignoring the role of data in process mining.Based on Petri net theory and association rule learning,thesis analyzes a large number of low-frequency behaviors and noises in event logs from the perspectives of control flow and data flow to judge whether they are abnormal behaviors.(1)By learning association rules among activities in the event log control flow structure,the set of activities with high support usually represents a high dependency between activities.However,the activity set with low support degree is regarded as lowfrequency behavior due to its low frequency,and the low-frequency behavior is often regarded as illegal behavior or abnormal behavior and deleted by most researchers,without exploring some internal links between activities.For the retained activity sequences with high incidence,the non-conforming activity sequences are eliminated by calculating their confidence.(2)after mining the association rules of the control flow structure of the event log,it is still necessary to consider the influence of the data flow(internal properties contained by the activity itself: resources,time stamps,etc.)on the judgment of abnormal behavior.In the same way,association rules are used to mine the data attributes of activities and deal with the anomaly of data attributes of activities.Finally,the process discovery algorithm is applied to the filtered event logs to mine the final process model.(3)Consistency checking is applied to evaluate the deviation between the model and the log.Aiming at the existing consistency checking methods,which are mostly based on control flow,a consistency checking method based on data flow is proposed.At the same time for Petri net can not describe the data in the role of the process model,put forward a kind of Petri net based on extended data-aware Petri nets,activity to use the complete presentation of data in a process model,and use the CRUD matrix to the use of the data in the process model specification,provide the basis for data flow alignment. |
