FDDPG: a federated deep deterministic policy gradient-based method for attack detection in IoV

To address the issues of data privacy leakage and the inability of models to adapt to dynamic network environments in Internet of Vehicles intrusion detection, this paper proposes an attack detection method based on Federated Deep Deterministic Policy Gradient. This method constructs a distributed privacy-preserving framework that protects data privacy through federated local model training. Subsequently, a DDPG intelligent agent mechanism is introduced to adaptively regulate model parameters, effectively resolving the convergence challenges caused by heterogeneous data. To further guarantee data security, the model integrates controlled Gaussian noise injection and gradient clipping mechanisms, realizing parameter updates that satisfy differential privacy requirements. Experimental results demonstrate that FDDPG outperforms baseline methods in terms of both accuracy and F1-score across three datasets, including UNSW-NB15. Under a balanced privacy configuration, the system can restrict accuracy loss to within 1% while ensuring strong privacy protection.