基于行为轮廓的二维傅里叶变换流程预测

2D discrete Fourier transform process prediction based on behavior profiles

现有的预测性流程监控大多数侧重于深度学习技术,很少有将流程中的行为与此结合.针对这一问题,结合行为关系提出了一个基于二维离散傅里叶变换的流程预测方法,对流程的下一个活动进行预测.将数据进行预处理,利用数据转换工程将时间数据转换为包括活动通道和性能通道的二维空间数据,对活动之间的行为关系进行活动行为编码,并将得到的矩阵进行二维离散傅里叶变换,输入CNN网络中进行训练并预测.使用仿真事件日志和真实事件日志进行评估,在仿真数据集和Helpdesk数据的测试集、BPIC12W数据的测试集上,本文方法预测准确度相比于CNN方法分别提高了2.57%、4.63%和1.67%.实验结果展示了本文方法能有效地提高流程预测的准确度.

Most of the existing predictive process monitoring focuses on deep learning techniques,and few combine the behavior in the process with this.In response to this issue,this article proposed a process prediction method based on two-dimensional discrete Fourier transform,which combined behavioral relationships to predict the next activity of the process.The method preprocessed the data,used data transformation engineering to convert the time data into two-dimensional spatial data including activity channels and performance channels.The method encoded the behavior relationships between activities and performed two-dimensional discrete Fourier transform on the obtained matrix.Finally,the method input it into the CNN network for training and prediction.This paper used simulation event logs and real event logs for evaluation.On the test sets of simulation dataset,Helpdesk data,and BPIC12W data,the prediction accuracy of this method improved by 2.57%,4.63%,and 1.67%compared to the CNN method,respectively.The experimental results demonstrated that the method proposed in this paper can effectively improve the accuracy of process prediction.