【作者】 裴岩；

【导师】 杨广明； 韩宇琮；

【作者基本信息】 东北大学 ， 软件工程， 2009， 硕士

【摘要】 机器学习在人工智能领域的研究中具有十分重要的地位。一个不具有学习能力的智能系统难以称得上是一个真正的智能系统,但是以往的智能系统,尤其是导航系统都普遍缺少学习的能力。机器学习在导航系统中的应用研究的目的是应用机器学习技术使车载导航系统实现智能化,提高导航产品的技术含量,所以课题的技术意义和社会意义都是十分深远的。在车辆导航系统中,实现对于系统规划路径和实际走行路径的比较和记忆的功能是当前导航领域中比较具有前瞻性的研究热点。本文应用强化学习理论,构建了一个在车辆导航系统中的强化学习模型,实现了比较和记忆实际走行路径的路径学习算法,在最优路径算法中考虑利用学习过的路径。本文在超图理论的基础上,给出了一种适合最优路径算法建模的矢量超图的数学描述,在此基础上定义了矢量超图对象模型,推广了超图理论在最优路径算法领域的建模应用,拓展了相关概念。在矢量超图对象模型的基础上,给出了多目标最优路径算法模型的执行步骤和关键参数的确定方法,讨论了k_λ参数、k~*参数的确定和多目标折中函数的构造和应用问题,并且在算法过程中考虑短时交通流预测信息和利用学习过的路径的功能。实际工程应用中,给出了一种权衡系统时空性能和最优解的算法收敛方法,并将本方法和模型在实际的车辆导航系统中得到应用。使得实现后的系统可以用来指导制定和实施某项交通管理计划,调节交通流量,以减缓可能出现的交通拥挤和危险隐患。随着现代交通管理模式由被动管理向主动诱导的转变,预测信息显得越来越重要。时间间隔不超过15分钟的短时交通流预测成为交通流研究的难点和热点。本文应用数据拟合回归技术,研究了车辆导航系统中的短时交通流预测,在数据拟合模型上,实现了短时交通流预测算法。利用本算法可以预测某条道路或某个交通走廊在未来几分钟内交通流状况的变化情况。更多还原

【Abstract】 Research on Machine Learning is an important problem in the artificial intelligence. An intelligence system without learning capability is not a really intelligence system, but the ancient intelligence system, especially the navigation system is lack of learning capability at large. The aim research on machine learning in the navigation system is to improve learning capability of the vehicle navigation system and the product’s intelligence degree using the machine learning technology, so the meaning of issue is far-reaching both in the technology and society.In the vehicle navigation system, it is a hot study point to implement comparation and memory of the system plan path and real driving path. In this paper, it uses the forcement learning theory to constructs a forcement learning modle in the vehicle navigation system, implement an algorithm of the path learning to compare and remember the system plan path and real driving path, and it is considered in the optimization path algorithm. It proposes a optimization path algorithm with the vector hyper graph description based on the hyper graph theory, defines the vector hyper graph object model, extends the application of the hyper graph theory in the optimization path algorithm area, develops the concept in this area. It proposes the method of steps and key parameters in the multi-aims optimization path algorithm, discusses the definition of the parameters, k_λand k~* and the construction and application of the multi-aims split the difference problem, also to consider the traffic information flowing forecast function and using the learning path function in the algorithm process. It gives a convergence method balance the time and memory space also the optimization path solution in the real engineer project. It is implemented to plan and actualize the traffic management plan, to adjust the traffic flowing, to reduce the traffic jam and dangerous that would happen.To forecast the traffic flowing is more important with the change in the passiviness mangement to initiative guidance in the model of the traffic management. It becomes to the study hot point that short time traffic information forecast less than 15 minutes. This paper uses the data matching return technology to research short time traffic information forecast in the vehicle navigation system, implements the algorithm of the short time traffic information forecast based on the data matching model. It can get the information of the traffic flowing in the several minutes in the future on some route.更多还原