Privacy-Preserving Collaborative Estimation for Networked Vehicles With Application to Collaborative Road Profile Estimation
Privacy-Preserving Collaborative Estimation for Networked Vehicles With Application to Collaborative Road Profile Estimation
复制标题
网络车辆隐私保护协同估计及其在协同道路轮廓估计中的应用
DOI：
10.1109/tits.2022.3154650
复制
发表时间：
2022-10
影响因子：
8.5
通讯作者：
Huan Gao;Zhaojian Li;Yongqiang Wang
中科院分区：
文献类型：
--
作者：
Huan Gao;Zhaojian Li;Yongqiang Wang
在智能车辆系统中，为了提高道路安全性、乘坐舒适性和燃油经济性，道路信息（如道路轮廓）已被广泛使用。然而，实际的挑战，如车辆的异质性，参数的不确定性和测量的可靠性，使得它非常困难的一个单一的车辆准确和可靠地估计这些信息。为了克服这些局限性，我们提出了一种新的基于学习的协作估计方法，通过融合来自网络车辆的信息。然而，这些车辆之间的信息交换所需的协同估计可能会泄露敏感信息，如个人车辆的身份，这构成了严重的隐私威胁。为了解决这个问题，我们提出了一个统一的隐私保护协作估计框架，该框架允许连接的车辆通过利用多个车辆穿越同一路段进行的连续测量来迭代地改进估计结果。协作估计方法系统地将隐私保护方案纳入估计设计，并利用估计动态来掩盖交换的信息。不同于修补传统的隐私机制，如差分隐私，这将损害算法的准确性或同态加密，这将导致沉重的通信/计算开销，动态启用的隐私保护不牺牲准确性或显着增加通信/计算开销。数值模拟证实了我们所提出的方法的有效性。
Road information such as road profile has been widely used in intelligent vehicle systems to improve road safety, ride comfort, and fuel economy. However, practical challenges, such as vehicle heterogeneity, parameter uncertainty, and measurement reliability, make it extremely difficult for a single vehicle to accurately and reliably estimate such information. To overcome these limitations, we propose a new learning-based collaborative estimation approach by fusing information from a fleet of networked vehicles. However, information exchange among these vehicles necessary for collaborative estimation may disclose sensitive information such as individual vehicle’s identity, which poses serious privacy threats. To address this issue, we propose a unified privacy-preserving collaborative estimation framework which allows connected vehicles to iteratively refine estimation results through exploiting sequential measurements made by multiple vehicles traversing the same road segment. The collaborative estimation approach systematically incorporates privacy-protection schemes into the estimation design and exploits estimation dynamics to obscure exchanged information. Different from patching conventional privacy mechanisms like differential privacy that will compromise algorithmic accuracy or homomorphic encryption that will incur heavy communication/computation overhead, the dynamics enabled privacy protection does not sacrifice accuracy or significantly increase communication/computation overhead. Numerical simulations confirm the effectiveness of our proposed approach.