Business Process Optimization And Task Prediction Based On Event Log

With the continuous development of artificial intelligence and the Internet of Things,companies with a large amount of data will occupy a greater advantage in the international market.At present,many companies have built various types of information systems,including a large number of event logs,but the logs contain relevant domain knowledge,which is difficult for ordinary managers to obtain useful information.Therefore,how to better maintain the business process in the fierce competitive environment,How to better analyze and predict business process event logs and provide decision support for enterprise related solutions is very important.However,for large-scale data such as event logs for business process prediction,traditional manual extraction will consume a lot of manpower and material resources,and the efficiency is very low.Therefore,it is necessary to use deep neural network or machine learning to predict business processes.This dissertation mainly uses process mining technology to process logs,and uses the relevant knowledge of Petri nets and behavior profiles and corresponding methods to construct the target log or event model,so that all events in the log can be reproduced in the model,and then analyzes and optimizes the mined model to obtain the optimal process model.In the research of business process prediction,this dissertation mainly uses the deep learning model to carry out the related research of business process task prediction methods,mainly including the research of task prediction for the next activity of business process and the interpretability research of business process prediction using the light GBM model.The specific research work is as follows:(1)In order to solve the problem of process modeling of multiple systems in real scenarios such as enterprises or schools,this dissertation adopts a Petri net process model mining algorithm based on subsystem merging to mine the required model,and applies it to the instance of freshmen’s enrollment process to verify the feasibility of the mining algorithm.At the same time,the mining model is combined with the freshmen’s orientation system in real scenarios,The corresponding optimization of the model makes the optimized model more reasonable in process and more efficient than the original model.(2)Secondly,aiming at the predictive business process monitoring problem of enterprises or schools and other actual scenarios,a business process activity prediction method based on prior knowledge and residual network is proposed.The existing methods mainly use the knowledge in the field of natural language processing,and rarely use the prior knowledge of the log itself.Therefore,this dissertation proposes a deep neural network model that combines the prior knowledge,attention mechanism and residual network to predict the next activity of the business process.Using computer vision technology and process mining technology,the event records are converted into image descriptions,and then used as input to train the residual neural network model based on prior knowledge.Finally,the process mining algorithm extracts prior knowledge from the event log to combine the model to improve the prediction quality of unknown events.The feasibility of the algorithm is proved by contrast experiments and ablation experiments.(3)Finally,a explicable prediction method for business process monitoring is proposed to solve the problem that the predictive business processes of general enterprises or schools adopt the black box method of deep learning,which is less interpretable.Different from using the deep learning model combined with the past process execution data to predict the ongoing process in the past,this dissertation proposes a explicable prediction running process example.First,use light GBM to train an excellent model,and then use the model to predict each activity in each track,record all probabilities,and draw a new student’s opening process model diagram with probability by combining the process model,Compared with the traditional prediction of the next activity task results,it is more intuitive,which is conducive to managers to make decisions in a timely manner.Figure [27] table [23] reference [89]...