A Study On Workflow Mining Based On Hybrid Genetic Algorithm

The aim of workflow mining is to extract information of the task execution traces in the event logs and establish the explicit workflow model. However, current workflow mining algorithms mostly use local strategy, which can not guarantee that a globally optimal workflow model is mined. In addition, these algorithms are not robust when logs contain noise. The genetic algorithms performing global search can overcome problems of sequence tasks, parallel tasks, choice tasks, non-free-choice constructs, loop constructs, hidden tasks, etc. And the problem of noise is naturally tackled by the genetic algorithms, but it would search a better workflow model after many iterations. The genetic algorithm is easily premature restraining as well as later period searching efficiency in the evolution to be low.In this paper, we propose an improved approach based one the hybrid genetic algorithm. This algorithm adopts tournament and elite retention strategy to carry on select, then it uses hybrid adaptive strategy carry on crossover and mutation, respectively introduces the idea of simulated annealing into crossover and mutation, and organically combines with the idea of multiple population parallel genetic evolution. Through the transfer, the population is graded. It can speed up the evolution speed while keep the stability of the good individual evolution.According to the character of genetic algorithm, we combined five criterions for evaluating the quality of mined mode in workflow mining area to evaluate the results of our algorithm. The five criterions are PFcomplete, Bp, Br, Sp and Sr.The improved algorithm has been implemented within the ProM, and the results show that this improved approach can search a better workflow model within a shorter time.