An Incentive Mechanism for Frozen Watermark-Based Federated Learning on Electric Vehicles Suspension

Federated learning for electric vehicle suspension (FLEVS) is a promising approach to balancing the trade-off between the limited battery capacity and the increasing demand for sensing camera devices in electric vehicles. However, existing FLEVS systems face critical challenges, including energy constraints, privacy risks, and intellectual property protection for car enterprise cloud (CEC) trained models. To address these issues, this paper proposes a frozen watermarking-based federated learning (FL) framework that protects both model integrity and privacy. The proposed approach integrates frozen watermarking into the global model, ensuring intellectual property security without degrading model accuracy in a continual learning setting. Additionally, we design an incentive mechanism based on a Stackelberg game between the CEC and mobile vehicles (MVs) to encourage participation in FLEVS while maintaining system efficiency. This mechanism derives the optimal information security strategy and determines the closed-form solution for the optimal reward function and local iteration requirements. Numerical results demonstrate that the proposed scheme enhances privacy protection and system sustainability while preserving model accuracy and efficiency.