Abstract: Distributed Cyber-Physical Systems（CPS） often encounter physical faults and cyber anomalies by fault propagation due to strong connectivity among the sub-systems. This paper presents a new data-driven framework for system-wide anomaly detection for such issues. The framework based on spatiotemporal feature extraction built on the concept of symbolic dynamics for discovering and representing the interactions among the sub-systems of a CPS, and then learn a system-wide model via Restricted Boltzmann Machine（RBM） use the extracted features. The results show that the framework can capture multiple nominal modes with one graphical model, and can used for anomaly detection.

Key words: Cyber-Physical Systems（CPS）, anomaly detection, Restricted Boltzmann Machine（RBM）, graphical model

摘要： 在分布式信息物理融合系统（CPS）中，由于各子系统间的强耦合性，常常会因为故障的传播导致整个系统的物理故障和网络异常。针对这一问题，提出了一种新的基于数据驱动的框架用于检测系统范围内的异常。该框架是用于发现和表征CPS各个子系统间相互作用的一种基于符号动力学的时空特征提取方案，并将提取的特征通过受限玻尔兹曼机（RBM）学习到一个系统级的模型。实验结果表明，该框架可以通过一个图模型捕获CPS的多模态，同时可用于异常检测。

关键词: 信息物理融合系统（CPS）, 异常检测, 受限玻尔兹曼机（RBM）, 图模型