Alignment-based Online Conformance Checking For Business Processes

Process mining represents an important research and industry topic,including modeling the complete operation process of an enterprise,extracting knowledge from event logs to build a real execution model,analyzing the relevant data of business processes to extract useful information and so on.With the realization of more and more process-aware information systems,more and more system-based data are recorded,so the demand of process mining related technology is also increasing.Since most information systems allow a certain amount of flexibility and deviation,conformance checking in process mining has become useful,and alignment has gradually become the main way to measure conformance.At present,there is few related work in the field of online process mining,and even less research in the sub-field of online conformance checking.The main purpose of online conformance research is to reduce the delay in finding deviations and to provide more timely and effective feedback to process decision makers.There are several different approaches to online conformance checking.The first one is to assume that all instances of the event flow are executed from the beginning of the process,and to calculate the conformance online by means of prefix-alignment.The second is to analyze the transition system,calculate all possible deviations in advance and add them to the transition system,and replay the event stream data on the transition system.The third is to construct the process model by using the behavior model and the weakly ordered relation semantics,and measure the consistency level by counting the number of effective and invalid behavior patterns.However,there are two shortcomings in both offline and online conformance checking.One is that most of the alignment costs are standard cost functions,resulting in the frequent occurrence of optimal alignment at equal low cost.The other is that it does not provide reliable comprehensive information about conformance.In conclusion,based on the alignment theory,this thesis proposes a global optimal alignment technology based on event log and the online conformance technology based on event stream.The main contents are as follows.(1)Aiming at dealing with the equal low-cost optimal alignment,the idea of global optimal alignment is given.When multiple same low-cost optimal allocations are available,the existing methods are mainly to change the cost function,either by maximizing synchronous movements and defining non-skippable activities,or by increasing the cost value of some movements and setting the cost of some movements to 0.However,the change of cost function can only reduce the emergence of equal low-cost optimal alignment to a certain extent,and does not fundamentally solve the problem.This thesis not only defines the adjustment cost function according to the severity of the deviation,but also proposes the principle of least repair costs to obtain the optimal alignment of multiple sets of logs,so as to select the global optimal alignment.(2)A new alignment method of truncated alignment is used to solve the warm start and complete stop phenomena that often occur in event streams.When conformance studies are conducted based on event flows,traditional alignment methods result in high deviation and low consistency results,while prefix alignment methods allow for complete termination but do not take into account warm start.This thesis not only defines truncated alignment,but also classifies transitions in the framework of the algorithm.For the example with sequential transitions as the starting event,the initial alignment mark is the enabling mark of the starting event,and for the example with concurrent transitions as the starting event,the initial alignment mark is the enabling identifier of the model.(3)The conformance of event flow alignment with the model is measured by referencing compliance,completeness and confidence.The commonly used conform-ance measures are for traditional alignment,without considering the particularity of truncated alignment or prefix alignment.Therefore,confidence is needed to judge the credibility of compliance to prevent the phenomenon of false high compliance.In this thesis,when judging whether the sequence of events is consistent with the model,we can't refer to a single index,but we need to compare the three indexes together to get a conclusion.(4)An algorithm framework of step-by-step optimal truncation alignment is proposed,which can calculate the optimal truncation alignment of the current example immediately every time a new event is received by the event stream,and at the same time,the optimal alignment obtained by other examples after the previous event is received is passed to the current moment.The algorithm framework of timely consistency detection and online consistency detection is proposed,which are suitable for two different application scenarios respectively.No matter which algorithm,when specifying a Petri net,the transition and activity matching table,the model reachable identification graph,the transition enabling identification,the model cost-free movement set,and the distance mapping function are obtained.Figure[18]Table[26]Reference[78]...