Intrusion Detection System in Industrial Cyber-Physical System Using Clustered Federated Learning

Intrusion Detection System in Industrial Cyber-Physical System Using Clustered Federated Learning

연구 분야: Infrastructure

학회: SN Computer Science

The quick convergence in the industrial infrastructure with computing and networking has proliferated the attacks of industry-based Cyber-Physical Systems (CPS). Protecting such massive, sophisticated, and varied industrial CPSs from cyber-attacks is difficult. This research proposed a novel Clustered Federated Learning (CFL) technique that identifies cyber-attacks against industrial CPSs. A CFL architecture enables many industrial CPS to create an extensive model for maintaining privacy. The contribution of the research is to identify the intrusion in the CPS using CFL that is highly effective in terms of accuracy, error rate, and MCC. CFL is a Federated Multi-Task Learning (FMTL) paradigm that groups the attacks into clusters with simultaneously trainable data distributions by taking advantage of geometric features of the Federated Learning (FL) loss surface. Extensive tests on an industrial CPS dataset show their proposed CFL technique's high efficacy in identifying cyber-attacks on industrial CPSs and its advantages over cutting-edge solutions. Experiments are conducted on gas pipelining system-based datasets. In this research study, the data resource is split into two main categories: training is 80%, and testing is 20%. The proposed CFL method is effective by attaining the highest accuracy of 99.3%, the minimal error rate of 86%, and the highest Mathews Coefficient Correlation (MCC) of 0.7994 for communication round 10.