Research And Implementation Of Incremental Process Prediction Method

With the development of information technology,enterprises and organizations store process information of business process execution in the form of event logs in their systems.Using process mining technology,enterprise managers can explore deeply from the event logs to identify the existing bottlenecks and take effective measures to improve and optimize business processes,thus increasing the efficiency of business process execution.Business process activity prediction is a hot topic of research in process mining.By accurately predicting the next activity of business process,it can assist relevant personnel to reasonably arrange activities and assist managers to reasonably allocate resources,thus making business process execution more efficient.However,with the development of social economy,the business processes of enterprises become more and more complex,the log of business process events becomes more and more huge,and the speed of business process changes becomes relatively faster.In addition,business processes are subject to process branching increments over time.Therefore,the traditional business process prediction based on historical event logs suffers from information lag in the online environment and is no longer applicable to the current complex and changing real-time online environment.Based on the above background,this paper investigates incremental process prediction methods in real-time event flow scenarios.The main work of this paper is as follows:(1)To address the problem of learning new event knowledge in online real-time scenarios,this paper proposes the incremental learning process prediction algorithm SWIL-LSTM based on incremental learning and LSTM to achieve dynamic process prediction in real-time scenarios,and solve the problem of degraded prediction performance of the model when new knowledge appears in the data stream through incremental learning.The results of experiment show that the business process activity real-time prediction algorithm based on incremental learning proposed in this paper can better adapt to real-time event flow scenarios,and improve the stability and accuracy of the business process prediction algorithm in real-time event flow scenarios.(2)To address the concept drift problem in online real-time scenarios,this paper further proposes the incremental learning process prediction algorithm DD-SWIL-LSTM incorporating concept drift detection,and uses concept drift detection to guide the incremental update of the prediction model,as a way to cope with the negative impact of outdated data weights on the prediction accuracy of the model.The experimental results show that the accuracy and stability of the business process prediction model prediction are improved under the guidance of the concept drift detection algorithm,and the proposed incremental learningbased business process activity prediction algorithm shows stronger adaptability in real-time prediction scenarios.(3)Based on the above proposed algorithm,this paper designs and implements a prototype system for real-time business process activity prediction,predicts the next activity for events in the real-time event stream and displays them on the page,and provides a display of historical accuracy and drift points to facilitate users to understand the credibility of the model prediction results,which provides a reference basis for the realistic application of this paper’s algorithm and reflects the effectiveness and Practicality.