【作者】 吴超仲；

【导师】 严新平；

【作者基本信息】 武汉理工大学 ， 载运工具运用工程， 2002， 博士

【摘要】 自动公路系统(Automated Highway System，简称AHS)是智能运输系统(Intelligent Transport System，简称ITS)领域中技术难度最大的系统。该系统包括车辆自动导航和控制、交通管理自动化以及事故处理自动化。这是智能运输系统的最终目标，即实现人、车、路高度一体化。自动公路系统的关键技术之一是车道保持系统，它是在道路导航设施和车载传感器等基础设施上，实现车辆自动行驶在期望的车道上。从车辆获取道路信息的方式来分类，自动驾驶有视觉导航和磁信号导航。视觉导航容易受天气、地面环境影响，不利于推广，而磁信号导航具有不受天气影响、可靠性高等优点，可以弥补视觉导航的不足，两者的联合使用可以提高安全性和可靠性，真正实现路-车一体化。磁性道钉导航是磁信号导航方式的一种，由于它具有成本低、易维护等优点，近几年成为国际研究的热点。国内在自动驾驶方面的研究主要集中在视觉导航方式上，以磁道钉方式导航的车道保持系统的研究除了本课题组外国内还没有见到相关报道。 本文以国家智能运输系统研究中心的自动公路系统试验车和专用的磁道钉道路为工作平台，在大量的现场实验基础上，针对基于磁道钉导航的车辆车道保持技术中的若干问题展开研究，运用神经网络、模糊逻辑、遗传算法等方法在多传感器信息融合、控制模型等方面进行了系统的研究，主要完成了以下几方面的工作： (1)对车道保持系统硬件平台进行了设计，论述了导航设施——磁道钉的设计过程和车载工控机、步进电机等的选择。着重讨论了磁传感器的设计过程，并从信号的幅值、信号的信噪比、同时性和对称性等方面对传感器进行了测试，证明了设计的磁传感器获取道路磁场信息是可以用于自动控制的，后面建立控制模型奠定了基础。 (2)通过分析信息融合技术的研究现状，针对自行设计的传感器获取的信武汉理工大学博士学位论文息源的特点，提出了三种信息融合方法，分别建立了数学模型。尤其对神经网络方法进行了深入的研究，建立了神经网络结构，确定了学习算法，并且通过专门的实验获取训练数据，得到了收敛的网络权值。最后，用实验比较了三种模型的优劣，确定神经网络信息融合方法比较接近实际。信息融合模型的建立，实际上是汽车横向位置的定位，为后面进行控制模型的研究奠定了基础。 (3)设计了基于磁道钉导航的模糊控制器，对遗传算法进行了研究，提出了变区域遗传算法，通过实例验证了该算法的有效性，并用于搜索模糊控制器中的参数，取得了较好的效果。现场实验结果表明，该模糊控制器可以实现自动控制，但控制精度有待于完善。 (4)从理论上探讨了磁道钉编码的意义，提出了磁道钉编码系统结构，并针对本次试验设计了磁道钉编码系统，现场试验结果表明，磁道钉编码系统是可靠的、实用的。 (5)在模糊控制的基础上，提出了模糊神经网络控制器，并用本文提出的变区域多层遗传算法与BP相结合进行网络学习，在理论上证明了遗传学习算法的收敛性。经过模型闭环系统训练，得到模糊神经网络控制器。用简化的二轮模型对本章设计的模糊神经网络控制器进行仿真，仿真结果表明，本文设计的模糊神经网络控制器比普通的模糊控制器在性能上有所改善，提高了控制模型的性能。 (6)在讨论了学习控制研究方法和水平的基础上，分析了自动公路系统中控制模型引入学习，尤其是在线学习的重要性和必要性。并设计了车道保持系统在线学习控制结构模型，为下一步实现真正的在线学习控制提供了理论基础。更多还原

【Abstract】 Automated Highway System (AHS) is the most difficulty sub system in Intelligent Transport System(ITS). The system includes vehicle automatic navigation and control, traffic management automation and traffic accident treatment automation. It is the final goal for ITS, i.e. integration of human, vehicle and road. Lane keeping system is one of the key technologies of AHS in which vehicle can be run following the anticipant lane automatically based upon the navigation installation and vehicle- mounted sensors. Vision navigation and magnetic information navigation are two types of automatic driving. Vision navigation is easy to disturbed by weather and road condition, however magnetic marker navigation hasn’t these disadvantages. Its safety and reliability can be enhanced if these two types are used in one system. This is why magnetic marker navigation becomes a hotspot in recent years internationally. In China, vision navigation is the main research project for automatic driving, however only our group has studied magnetic marker navigation in lane keeping system, no any other reports have been searched.This dissertation outlines the study on several problems of vehicle lane keeping system based on magnetic markers. Site experiment were carried out on the testing vehicle for AHS and the magnetic markers road in National Intelligent Transport System Center. Also, study on multi-sensor information fusion and control model by using neural net system, fuzzy logic and genetic algorithm were carried out. The work completed are as follows mainly:(1) Hardware system for lane keeping system including magnetic marker, magnetic sensor, control system and executor machine were designed. Research progress of magnetic sensor is described in detail. Test result indicated that the self-designed magnetic marker can be used to obtained magnetic information.(2) Having analyzed the information fusion technology, threemathematical models were made based on the signal sources obtained from self-designed magnetic sensor. The neural net method is discussed in detail, the configuration of neural net is established, study algorithm is decided. Training data were obtained from experiment. Simulation indicated that neural net method is more accurate than’the others two methods. In fact, information fusion modal is the location of vehicle lateral position which establishes research fundament of control modal later.(3) The fuzzy controller based on magnetic markers is designed. The changeable area and hierarchical genetic algorithm is studied; several examples indicated that this algorithm which can be used to search parameter of fuzzy controller is effective. Site experiment showed that designed fuzzy controller can be used in lane keeping system.(4) The significant of magnetic marker coding system is approachedtheoretically, structure of this coding system is introduced; this coding systemis designed according to test requirement.(5)Based on fuzzy control theory, fuzzy neural network (FNN) controlleris introduced; the changeable area and hierarchical genetic algorithm is usedto training fuzzy neural network. Simulation indicated that this designedfuzzy neural controller is effective.(6) Based on the discussion on learning control method, control model learning in AHS, particular in the important and necessity of online learning, is analyzed. Structure model of online learning control for lane keeping system is designed. It provides the theoretical fundament for further study on online learning control in the future.更多还原