《分层级联信息融合车辆定位及动态路径规划研究》由会员分享，可在线阅读，更多相关《分层级联信息融合车辆定位及动态路径规划研究（152页珍藏版）》请在金锄头文库上搜索。
1、湖南大学 博士学位论文 分层级联信息融合车辆定位及动态路径规划研究 姓名：胡林 申请学位级别：博士 专业：车辆工程 指导教师：谷正气 20071216 博士学位论文 - II - 摘 要 近几年来，我国汽车拥有量以 15%以上的速度递增，而公路里程的增长率只 有 3%左右，目前中国每千人拥有不到 50 辆汽车，全球平均水平为 120 辆，美国 为 750 辆，中国消费者对汽车还有着巨大需求，在未来相当长一段时间里，中国 汽车市场将继续保持快速发展，因此车与路的矛盾日益突出。ITS 是解决这一矛 盾的有效手段，车辆定位导航系统是 ITS 的重要组成部分，可以为出行者提供实 时定位，进行最优路径规
2、划，优化城市交通流，提高运输效率。近年来，许多研 究机构都在十分活跃地研究车辆定位导航领域的各种关键技术，以提高车辆定位 导航系统的性能、减低成本和解决实际应用中的问题，但车辆定位导航系统在实 际应用和测试中表现出了许多不足。在这种情况下，本文围绕车辆定位与路径规 划这两个车辆定位导航系统的关键技术进行了深入研究，以学术意义和实用价值 并重为原则，在车辆定位、地图匹配方法、信息融合技术、动态路径规划等方面 取得了一定成果。最后自主开发了一个低成本的车辆定位导航系统。本文工作的 主要贡献和创新总结如下： 1.提出基于改进卡尔曼滤波的传感器融合的车辆定位技术：从成本、可靠性 和精度综合考虑，采用两
3、个光学编码器作为当地传感器，一个廉价的 GPS 接收器 作为全球传感器，GPS 和编码器组合成一对互补的传感器，对车辆来进行高精度 的定位。传感器融合软件由 2 部分组成：一是基于车辆运动学的动态模型，它用 于结合 GPS 和光学编码器， 另一部分是用一个改进的卡尔曼滤波器来整合随时间 变化的传感器数据。为了验证滤波器的滤波效果，进行了仿真和实验，其结果表 明本文提出的定位系统效果很好，可实时地精确地预测车辆位置和方位角，GPS 信号中混杂的噪声可消除 80%，同时成本也很低。 2.设计了基于改进 D-S 证据的地图匹配方法：由于地图匹配中传统 D-S 证据 理论算法存在判别信息之间的相互关系
4、时过于绝对化和经验化的问题，我们对此 进行了改进，采用 Pignistic 概率的决策理论和基于权值 D-S 证据推理的地图匹配 新算法，建立车辆位置信息和方向信息判决规则，有效地利用现有信息扩大各匹 配道路之间的信度差异，提高了算法的鲁棒性。解决了证据的可信度和决策规则 问题，使新的组合推理规则能够有效地处理证据中的冲突信息。充分利用导航系 统的位置和方向信息，能够根据道路网的特点，选择适当的方向和位置可靠性参 数，利用现有信息有效扩大各假设命题之间的信度差异，在不提高决策风险的前 提下，使得决策更准确有效。 3.基于 ANFIS 的车辆定位系统信息融合的技术： 采用一种分层级联的信息融 分
5、层级联信息融合车辆定位及动态路径规划研究 - III - 合结构，对车辆定位与路径规划系统中应用不同导航定位技术得到的定位信息进 行融合，构造了应用于决策级融合的 ANFIS 结构，讨论了实际应用中 ANFIS 的 初始设置和训练，采用实际跑车数据对提出的方案进行试验验证，并与采用另外 两种决策级融合方法的结果进行了比较分析。 4.对城市间动态路径规划提出了混合遗传算法：利用人类对道路网络和路径 搜索过程的认知，引入了分层算法，缩减搜索规模，对城市间路网以改进 A*算法 来产生初始种群，利用 GA 的全局搜索和-交换局部搜索技术的局部搜索性能， 大大加快了对最优路径的求解速度。 5.在城市内路
6、径规划中考虑了路口交通信号所导致的延时，并首次提出了一 种标签修正新算法ARSC 算法：对城市内路网，考虑十字路口交通信号的延 时，对信号控制规律已知的 STV 路网，基于 LET 适应路径策略，提出了惩罚函 数算法；对于信号规律不确定，只知道其概率的情况以及由于错过时间而造成不 可预测的等待的情况下，提出了一种标签修正算法ARSC 算法，它是对 ELB 算法的扩展，明确地考虑了由于信号控制而导致的附加延迟，且只需要概率性地 了解信号规律和延迟时间。 关键词：车辆定位；路径规划；传感器融合；卡尔曼滤波；地图匹配；ANFIS； 分层算法；遗传算法；惩罚函数；ELB 算法；ARSC 算法 博士学位
7、论文 - IV - Abstract In recent years, the vehicle population of China has increased by 15%, but the growth rate of total mileage is only about 3%. Nowadays in China, the number of vehicles per 1000 population is lower than 50, compared with the world average level over 120 and the number in America 75
8、0, there is great demand for vehicles in China market, so the China vehicle market will keep fast development for a long period. In such a case, the significant gap between the volume of traffic and the capacity of road will be increasingly serious. ITS would be a effective method to solve this prob
9、lem, the vehicle location and navigation systems, as an important component of ITS, can provide motorists with real-time route guidance, optimal path planning, and then optimize the urban traffic flux and improve the transport efficiency. Although there are many research productions have been achiev
10、ed in this research field, even some study results have became actual products, a lot of dissatisfactory facts are still existed. In order to enhance the performance of vehicle location and navigation system, advance its practical problem-solving ability and reduce the cost, many research institutes
11、 have been working hard on the key techniques of vehicle location and navigation in recent years. In order to effectively carry out the development research work of vehicle location and navigation system, based on the principia of science significance equal to the practical value, this thesis lucubr
12、ates the two key techniques of vehicle navigation system, and make a series of innovation fruits on the researches of map matching, information fusion of vehicle location, dynamic route optimal technique and so on. Finally an independent innovative low-cost vehicle location and navigation system pro
13、totype was made. The major work and innovation in this thesis can be summarized as the following: 1.Propose the vehicle location technology based on improved Kalman filter sensor fusion. From the points of view of cost, reliability and precision, two optical encoders were used as local sensors, and
14、a low-cost GPS receiver worked as a global sensor. Then the GPS and encoders were integrated to form a complementary sensor pair for high accuracy orientation. The software for sensor fusion was composed of two parts: one innovative part is based on the vehicles kinematics model to integrate GPS and
15、 optical encoders; and another part is a time-varying Kalman filter to merge 分层级联信息融合车辆定位及动态路径规划研究 - V - sensor data. Both simulations and experiments were carried out to investigate the effects of filtering noise. The results showed that the orientation system did a good job, It could accurately es
16、timate vehicle position and heading angle in real time. The noise mixed with the GPS signal was reduced by 80% with filtering for practical applications and the cost is relative lower. 2.Develop the map-matching algorithm based on weight D-S evidence. The traditional D-S evidence theory algorithm has some shortcoming in judging the relationship among each information, say the absolutization and empiricism. In this thesis it is improved to solve the problems about the credibility of evide
