Scholarlink

<p>In the dynamic landscape of sustainable transportation, addressing privacy concerns in electric vehicles (EVs) battery state of health (SoH) prediction is vital yet challenging. We identify various scenarios highlighting security vulnerabilities from the perspectives of EV owners/users and manufacturers/service providers. To mitigate these concerns, our proposed methodology integrates homomorphic encryption(HE)-based privacy-preserving techniques with a deep neural network (DNN) processing framework to securely handle sensitive user data intrinsic to EV systems, including location, usage patterns, and personal preferences. The proposed HE-enabled Privacy-Preserving Deep Neural Network (PPDNN) Framework can provide privacy in the training and inference phases with training data privacy, model privacy, input privacy, and output privacy. A comprehensive security analysis demonstrates the system’s effectiveness against various attacks throughout the machine learning stages. Experimental validation utilizing the Tenseal package on real battery cell datasets (B0005, B0006, B0007, and B0018) from NASA repository demonstrates the performance of both DNN and PPDNN models. The results show a decreasing mean absolute error (MAE) trend with increasing training epochs, indicating effective learning and convergence. Despite training on encrypted data to ensure privacy, the PPDNN models closely match the performance of DNN models, underscoring the HE-PPDNN framework’s ability to maintain accuracy while preserving privacy effectively.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

PPDNN-SoH: privacy-preserving deep neural network for sustainable electric vehicle battery state of health prediction

  • 期刊论文

  • 发表日期：

    2025年9月28日

查看全文
  • Vankamamidi S. Naresh,
  • Vanapalli Sai Sriram

摘要

In the dynamic landscape of sustainable transportation, addressing privacy concerns in electric vehicles (EVs) battery state of health (SoH) prediction is vital yet challenging. We identify various scenarios highlighting security vulnerabilities from the perspectives of EV owners/users and manufacturers/service providers. To mitigate these concerns, our proposed methodology integrates homomorphic encryption(HE)-based privacy-preserving techniques with a deep neural network (DNN) processing framework to securely handle sensitive user data intrinsic to EV systems, including location, usage patterns, and personal preferences. The proposed HE-enabled Privacy-Preserving Deep Neural Network (PPDNN) Framework can provide privacy in the training and inference phases with training data privacy, model privacy, input privacy, and output privacy. A comprehensive security analysis demonstrates the system’s effectiveness against various attacks throughout the machine learning stages. Experimental validation utilizing the Tenseal package on real battery cell datasets (B0005, B0006, B0007, and B0018) from NASA repository demonstrates the performance of both DNN and PPDNN models. The results show a decreasing mean absolute error (MAE) trend with increasing training epochs, indicating effective learning and convergence. Despite training on encrypted data to ensure privacy, the PPDNN models closely match the performance of DNN models, underscoring the HE-PPDNN framework’s ability to maintain accuracy while preserving privacy effectively.