ADP-VFL: An Adaptive Differential Privacy Scheme for VPP Based on Federated Learning

In recent years, with the remarkable development of Virtual Power Plants (VPP) and the surge in the number of Electric Vehicles (EVs), the issue of data privacy leakage has become increasingly prominent. The effectiveness of existing federated learning schemes in mitigating data privacy leakage, it still faces potential threats such as inference attacks and user and server collusion. To protect the privacy of federated learning, some schemes have introduced differential privacy(DP). Nevertheless, applying DP will inevitably affect the accuracy to some extent. In this paper, we propose an adaptive differential privacy scheme for VPP based on federated learning, named ADP-VFL. Our ADP-VFL scheme can achieve data privacy preservation by transmitting the noise-added data as a chain, defend against inference attacks by innovating offset noise mechanism and a parallel transmission scheme. The performance evaluation results demonstrate that the proposed scheme can improve the aggregation accuracy and reduces the communication overhead.