FED-FAN: Federated learning-based intrusion detection system for Flying Ad-hoc Networks

초록

The increasing adoption of Unmanned Aerial Vehicle (UAV) swarms in applications such as disaster response, environmental monitoring, military reconnaissance, and smart transportation has exposed Internet-of-Drones (IoD) networks to sophisticated cyber threats, particularly Denial-of-Service (DoS) attacks. Traditional centralized intrusion detection systems suffer from privacy concerns, single points of failure, and limited scalability, making them unsuitable for distributed UAV environments. To address these challenges, this paper proposes FED-FAN, a blockchain-assisted federated learning framework for secure and privacy-preserving intrusion detection in Flying Ad Hoc Networks (FANETs). FED-FAN integrates FedProx-based federated learning with a lightweight CNN-LSTM intrusion detection model to enable collaborative attack detection without sharing raw data. To ensure tamper-resistant security auditing, the framework incorporates the lightweight PureChain blockchain utilizing the Proof of Authority and Association (PoA(2)) consensus mechanism for immutable attack logging. To emulate realistic UAV swarm environments, heterogeneous non-IID data distributions are generated using Dirichlet partitioning (alpha = 0.3), and extensive experiments are conducted on the WSN-DS and 5G-NIDD datasets. Furthermore, scalability and robustness analyses are performed using the UAV-NIDD and UAVIDS-2025 datasets across varying client populations. Experimental results demonstrate that FED-FAN achieves an accuracy of 96.54% on WSN-DS and 98.76% on 5G-NIDD under optimal client configurations while maintaining low loss values and stable convergence under heterogeneous data conditions. Additional evaluations confirm the framework's scalability and effectiveness across diverse UAV intrusion detection scenarios. The proposed framework contributes a dual-layer defense architecture capable of protecting both wireless sensor network communications and 5G-enabled UAV communications while providing verifiable blockchain-based attack auditing. These findings demonstrate the potential of combining federated learning and lightweight blockchain technologies to enhance the security, privacy, scalability, and trustworthiness of next-generation UAV swarm networks.

키워드