: Privacy-Preserving Prediction of Real-Time Energy Demands in EV Charging Networks
Beibei Li,Yuqing Guo,Qingyun Du,Ziqing Zhu,Xiaohui Li,Rongxing Lu
DOI: https://doi.org/10.1109/tii.2022.3182972
IF: 12.3
2023-03-11
IEEE Transactions on Industrial Informatics
Abstract:Real-time and accurate prediction of charging pile energy demands in electric vehicle (EV) charging networks contributes significantly to load shedding and energy conservation. However, existing methods usually suffer from either data privacy leakage problems or heavy communication overheads. In this article, we propose a novel blockchain-based personalized federated deep learning scheme, coined , for privacy-preserving energy demands prediction in EV charging networks. Specifically, we first design an accurate deep learning-based energy demands prediction model for charging piles, by making use of the CNN, BiLSTM, and attention mechanism. Second, we develop a blockchain-based hierarchical and personalized federated learning framework with a consensus committee, allowing charging piles to collectively establish a comprehensive energy demands prediction model in a low-latency and privacy-preserving way. Last, a CKKS cryptosystem based secure communication protocol is crafted to guarantee the confidentiality of model parameters while model training. Extensive experiments on two real-world datasets demonstrate the superiorities of the proposed scheme in accurately predicting real-time energy demands over state-of-the-art schemes. Further, the scheme can achieve reasonably low computational costs, compared with other homomorphic-based schemes, such as Paillier and BFV.
automation & control systems,computer science, interdisciplinary applications,engineering, industrial