Privacy-Preserving Smart Energy Management by Consumer-Electronic Chips and Federated Learning
Huakun Huang, Sihui Xue, Lingjun Zhao, Weizheng Wang, Huijun Wu
Abstract
With consumer electronics (CE) development, green Internet-of-Vehicles (IoV)-based renewable energy systems have attracted an upsurge in interest. Nonetheless, efficient energy management and high renewable energy efficiency are facing crucial challenges. To tackle such issues, motivated by the virtual power plant (VPP), we propose a smart energy management scheme, integrating the merits of software and hardware. Particularly, leveraging electrical vehicles (EVs) and reconfigurable CE chips, the proposed system enables EVs to autonomously charge or discharge batteries for intelligent storing of surplus electricity generated by renewable resources. However, such a system requires frequent sharing of sensitive data (i.e., EV users’ identification information, locations, etc.) between the control center and CE chips, resulting in privacy issues and communication overhead. For this, we propose a federal EV decision learning (FEVDL) approach. FEVDL allows EVs to share trained models without revealing sensitive data. FEVDL can achieve a competitive accuracy of about 99% compared with isolated edge learning. Meanwhile, it separately improves inference accuracy by about 4%, 15%, and 25% under three challenging conditions. Therefore, a privacy-preserving green IoV-VPP system can efficiently operate the distributed EV batteries as a large-scale power-storage facility.