不依赖于精确初始坐标的车联网相对定位坐标估计算法

Exact Initial Coordinate Free Relative Localization Algorithm for Vehicular Ad Hoc Network

在车联网定位中,GPS(Global Positioning System)信号长时间较差甚至中断会导致GPS定位结果不可靠甚至不可用,无法为相对定位算法提供可靠的精确初始坐标.针对这一问题,该文对车辆相对位置坐标估计方法展开研究,结合TOA(Time of Arrival)测距技术,将相对位置坐标估计问题转化为非线性规划问题.为减小非线性规划问题中初始坐标对算法结果的影响,将外部罚函数法与Powell算法结合,利用外部罚函数法"能够从非可行解出发逐步逼近可行域"的特点优化最优化方法,解决算法对初始坐标的敏感特性;利用Powell算法能够"逼近局部最优解"的特点作为最优化求解方法,用于求解目标函数最优解.提出一种不依赖于精确初始坐标的相对定位(Exact Initial Coordinate Free Relative Localization,EICFRL)算法,实现车联网高精度相对定位.在算法验证中,该文采用两种TOA节点部署方案,分别为单点部署方案和基于几何约束的多点部署方案.在多点部署方案中,利用车辆固有形状属性,形成基于车型的几何约束,增加非线性规划问题可行域限制.为验证该文算法可行性及有效性,该文在仿真实验中设置不同测距误差、连通性、车辆数目等条件,并在实际环境中实验验证,将该算法结果与Powell算法、LM(Levenberg-Marquard)算法、CRLB(Cramer-Rao Lower Bound)进行对比.实验结果显示,该文算法定位精度提高超过50%.当使用多点部署方案时,算法定位误差进一步减小约为30%(仿真环境)和23%(实测环境).

For localization in vehicular ad hoc network, GPS signals are easily disturbed or blocked by obstacles. During long-duration GPS outages, GPS position coordinates of vehicles will exceed acceptable levels of accuracy. In this paper, the relative location estimation problem was considered and transformed into a nonlinear programming problem with TOA. The exterior Penalty Function, approaching the boundary of the feasible region without initial feasible solution required, was used to minimize the sensitivity of algorithm to the initial coordinates. Combining with exterior Penalty Function, the corresponding optimization problem was solved by Powelltechnique. An Exact Initial Coordinate Free Relative Localization （EICFRL） algorithm was proposed for the relative vehicular localization. Two different deployment scheme with one or more TOA nodes were used in this work. In the latter one, geometric constraint, derive from inherent shape-based attributes, was exploited to restrict position relationship of nodes on each vehicle. As a result, this can lead to a better and more accurate tightening process of the feasible region. Feasibility and effectiveness of the proposed EICFRL algorithm was validated by field tests and simulations under variables including range error, connectivity and number of vehicles. Powell, LM and CRLB have been used for comparison with EICFRL algorithm. Experimental results demonstrate that our algorithm achieves exceeding 50% increase in positioning accuracy. In addition, localization error is further reduced about 30% （simulation environment） and 23% （measured environment） using geometric constraint deployment.