Research On Process Mining Methods Based On Petri Net And Event Logs

Process mining is an emerging subject to connect data mining techniques with process modeling method, given the observed behavior recorded by information system, we can extract useful information from them and the process model is established automatically to describe the behavior using the process mining techniques. In the circumstance of information technology develop rapidly, business process management system plays an increasingly important role in the enterprise, and information system will record a large number of log files in the process of running, so we can restore operation process using the process mining techniques, which helps enterprises to improve business processes. Therefore, the research of process mining methods has important theoretical value and application value.Previous process mining methods based on the event log are mostly according to the causal relationship between tasks, and mainly based on direct dependencies between tasks to build process model. They have a lot of limitations. Although there are many methods can discovery indirect dependencies, they do not from the perspective of the process behavior to analysis. Behavior profile relationship can describe the order of activities well, so we can describe the causal relationship between tasks with behavior profile relationship between tasks. In addition, when the number of tasks or the traces of event logs is too large, the process of process mining is time-consuming. Therefore, how quickly and efficiently to mining process model is a problem worthy of studying. According to these two problems, two kinds of process mining methods are proposed based on Petri net and event logs. They can solve the two problems effectively after verification. In this paper, the main contributions are as follows:(1) There are few can mining process model containing indirectly dependencies among existing process mining methods. We use behavior profile relation to represent indirect dependencies between tasks in this paper, and the concept of quasi indirectly dependencies is given, so we puts forward an effective process mining method which can mining process model containing indirectly dependencies.The process mining algorithm based on quasi indirect dependencies take the behavioral profiles into account and the initial model is established according to the behavioral profiles. Then adjusting model based on incremental logs and quasi indirect dependence. Finally, selecting the optimal model according to the evaluation criteria. This method is especially suitable for mining the process model with indirect dependence.(2) When the number of tasks or the traces of event logs is too large, the process of process mining is time-consuming. In order to solve this problem, a new process mining method is proposed based on model merged. Firstly, sub-models are established respectively according to the behavioral profiles of each event log. Then finding out the maximum matched regions of the sub-models, merging the maximum matched regions and adding the rest of the transitions into the suitable place. Thus, the merged model is obtained. The method considers the maximum matched regions as a whole, greatly simplifying the process of modeling. Finally, a simple example is listed to show the feasibility of the method.