Learning-Based Online Transmission Path Selection for Secure Estimation in Edge Computing Systems

Abstract

Edge computing is emerged as a promising solution to cope with huge volumes of data generated by smart devices and low latency demand for mission-critical applications in industrial cyber–physical systems. Data processing and estimation are shifted to the edge computing side. Nevertheless, the WCN between field devices and edge computing side is exposed to malicious attackers because of its openness. Therefore, in this article, we focus on the transmission path selection strategy design to guarantee the secure state estimation on the edge side against dynamic denial-of-service attacks. First, we present a novel learning-based secure routing algorithm (LSRA) to learn the attack rule and predict the attacker's next conduct with the use of both historical and online data. With lower computational complexity, the proposed learning algorithm could track the attack rule in real time whenever new data comes. Meanwhile, we derive the analytical relationship between the probability upper bound of learning error and the learning time. Based on the predicted attacker's behavior obtained by the learning algorithm, we flexibly select the secure routing path to avoid being attacked and, thus, improve successful transmission probability. Furthermore, this secure routing path selection method improves the performance of the state estimation system. The theoretical analysis of estimator stability is given. Finally, simulation results reveal the effectiveness of LSRA and the path selection scheme.