Federated reinforcement learning–driven multi-task optimization for robust and ethical edge internet of things security

Campus security systems empowered by edge Internet of Things (IoT) technologies face inherent conflicts among detection accuracy, resource efficiency, privacy protection, and adversarial robustness. Traditional Federated Learning (FL) frameworks and quantization methods often struggle to achieve balanced optimization under heterogeneous device conditions and strict ethical constraints. To address these challenges, this study proposes a Federated Reinforcement Learning (FRL)-driven multi-task collaborative optimization framework for secure and efficient edge IoT environments. The framework constructs a hierarchical perception-decision-constraint architecture that enables dynamic adaptation of quantization precision, defense intensity, and privacy parameters based on real-time environmental states. Ethical compliance is continuously ensured through distributed auditing and automated constraint enforcement, forming a closed-loop optimization mechanism integrating technical robustness and normative governance. Extensive experiments conducted on the NSL-KDD benchmark demonstrate that the proposed system achieves 94.3% intrusion detection accuracy across twelve heterogeneous nodes, with an F1-score of 89.3% for rare User-to-Root (U2R) attacks. The approach reduces energy consumption by 66.1%, latency by 72.8%, and limits adversarial attack success rate to 23.4%. Under Differential Privacy (DP) constraints with ε = 2.0, privacy leakage incidents are controlled within 2.4 occurrences per month, and the framework maintains stability over 168 h of continuous operation. Overall, this research provides an integrated, scalable, and ethically accountable solution for campus network security, demonstrating the potential of FRL to achieve synergistic optimization across performance, robustness, and compliance dimensions in edge intelligent systems.