Federated Learning-Enabled Security Enhancement for Distributed Autonomous Control Systems Against Malicious Attacks
Abstract
Distributed Autonomous Control Systems (DACs) are widely used in safety-critical fields like intelligent transportation and industrial automation, yet face growing threats from Byzantine attacks, data poisoning and jamming that may cause catastrophic failures. Federated learning (FL) addresses DACs’ privacy and communication issues but lacks dedicated security mechanisms for its training and deployment phases. This paper proposes the Federated Secure Learning Framework (FSLF), integrating Byzantine-resilient aggregation, attack-aware adversarial training and cryptographic communication validation to balance security and privacy. It filters malicious model updates, generates attack-specific perturbations for robust training and detects tampered communication data. Experiments on multi-UAV tracking, CAV platoon control and robot manipulation show FSLF achieves 92.3% Byzantine attack detection rate, 15.7% lower Avg-RMSE under attacks and 0.6% data leakage rate, boosting the security and reliability of FL-enabled DACs in adversarial environments.
