关键词: 异常检测, 上下文感知, 业务流程管理, 重演, 神经网络

Abstract: For anomaly detection,most existing methods only detect anomalies in business process executions from a single perspective.Therefore,a multi-perspective online anomaly detection method was proposed,which detected three kinds of anomalies that might appear in current execution simultaneously such as behavior anomaly,time anomaly and attribute anomaly.In addition,the replay technology was applied to fully extract the behavior contexts and data contexts from event logs to simulate the true environment of process executions and combine them with deep learning methods to build a real-time multi-perspective anomaly detection model.Experiments on four real-life datasets demonstrated that the proposed method vastly outperformed other approaches and increased the mean recall by 18.5%,19.8% and 8.4% and the mean F1 score by 15.9%,25.4% and 8.4% compared to Support Vector Machine (SVM),K-Nearest Neighbor (KNN) and Digital Twin (DT).Compared to traditional method that detected anomaly based on offline event logs,the proposed method was more suitable for online anomaly detection of the on going instance.

Key words: anomaly detection, context awareness, business process management, replay, neural networks