FSL: federated sequential learning-based cyberattack detection for Industrial Internet of Things

Abstract Industrial Internet of Things (IIoT) brings revolutionary technical supports to modern industries. However, today’s IIoT still faces the challenges of modeling varying time-series in common data isolation while considering data security. To accurately characterize industrial dynamics, we propose a possible solution based on federated sequence learning (FSL) with cyber attack detection capabilities. Under a federated framework, FSL constructs a collaborative global model without violating local data integrity. Taking advantages of the locally sequential modeling, FSL captures the intrinsic industrial time-series responses. Furthermore, data heterogeneity among distributed clients is also considered, which is important to maintenance a robust but sensitive attack detection. Experiments on classic distributed datasets demonstrate that FSL is capable to accurately model data heterogeneity caused by data isolation and dynamics of time-series. Real IIoT attack detection experiments using a distributed testbed show that our FSL provides better detection performances for industrial time-series sensory data compared to existing methods. Therefore, the proposed attack detection approach FSL is promising in real IIoT scenarios in terms of feasibility, robustness and accuracy.