摘要
现有的基于深度学习的业务流程剩余时间预测方法大多采用传统的长短期记忆循环神经网络构建预测模型,由于传统长短期记忆循环神经网络对序列数据的建模能力有限,导致现有方法的预测效果还有较大提升空间。针对现有方法的不足,提出一个基于注意力双向循环神经网络的业务流程剩余时间预测方法。该方法使用双向循环神经网络对流程实例数据进行建模,同时引入注意力机制自动学习流程实例中不同事件的权重。此外,为了进一步提升学习效果,基于迁移学习的思想设计了一种迭代学习策略,为不同长度的流程实例分别构建剩余时间预测模型,提高了模型的针对性。实验结果表明,所提方法与传统的方法相比具有明显的优势。
Most of existing deep learning approaches to business process remaining time prediction are used the traditional Long-Short Term Memory(LSTM)neural networks.Due to the limited capacity of modeling temporal sequential data,there is still much room for improvement in predicting accuracy.For the deficiency,the business process remaining time prediction method based on leverage bidirectional recurrent neural networks was proposed.An attention mechanism to automatically learn the weights of each event was also introduced in a business process instances.To tackle the variety among business process instances of different lengths,a transfer learning mechanism to learn prediction models for ongoing process instances was designed with a particular length.Extensive experiments showed that the proposed approach outperformed existing process-model-based approaches and LSTM-based approaches.
作者
倪维健
孙宇健
刘彤
曾庆田
刘聪
NI Weijian;SUN Yujian;LIU Tong;ZENG Qingtian;LIU Cong(College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266510,China)
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2020年第6期1564-1572,共9页
Computer Integrated Manufacturing Systems
基金
国家自然科学基金资助项目(61602278,71704096)
中国博士后科学基金资助项目(2014M561949)
青岛市社会科学规划研究资助项目(QDSKL1801122)。
关键词
剩余时间预测
业务流程
双向循环神经网络
注意力机制
remaining time prediction
business process
bidirectional recurrent neural networks
attention mechanism
