交通流预测与配流的研究

【作者】 丁浩；

【导师】 谭国真；

【作者基本信息】 大连理工大学 ， 计算机应用技术， 2003， 硕士

【摘要】 目前，作为解决交通拥挤问题最适宜的方法————交通诱导及交通配流成为智能交通系统研究的关键问题。本文将神经网络应用到交通流预测中。在分析传统神经网络缺陷基础上，运用线性独立函数构建了智能神经元模型，并对这种神经元的知识存储能力进行了理论分析。分析表明这种智能神经元较普通神经元具有更高的知识存储能力，从而使整个神经网络的信息处理能力得到大大增强。文中将这种智能神经元组成的神经网络应用于交通流预测，对大连市实测交通流数据进行了预测分析。实验结果表明由这种智能神经元组成的神经网络具有更高的收敛速度和精度，预测效果远远优于常规BP网络，达到实测路况交通流预测所需精度及收敛速度，具有较高实用价值。文中还给出了行程时间推算和路径优化的方法，提供了交通诱导的全套解决方案。 在智能交通系统研究的推动下，使得动态交通分配理论成为研究热点。作为一种新的全局优化搜索算法，遗传算法得到广泛应用。在上述理论基础上，实现了最优控制动态交通分配的主从式并行遗传算法，并在基于分布式存储和消息传递体系结构的曙光机群系统下，应用并行编程环境PVM下master／slave模式完成了上述算法。文中还应用具体网络算例测试了该优化模型，取得了较好的系统运行指标和并行效率。 网格计算的目标是将跨地域的多台高性能计算机、大型数据库、科研设备，通信设备以及其他潜在的计算资源整合成一个巨大的超级计算机系统，支持科学计算和科学研究。本文在介绍网格计算基本理论基础上，在现有硬件环境下架构了Globus试验平台，并在Globus网格环境下测试了最短路径问题的并行Dijkstra算法。同时建立了基于理工大学现有硬件环境的计算网格模型；在大连市原有交通控制系统基础上，描述了基于网格环境的交通控制系统的构建，并在该交通控制网格系统基础上，分析讨论了交通流预测在网格环境下的实现。更多还原

【Abstract】 By analyzing limitation of the traditional Neural Network, this paper presents intelligent neuron model based on linear independently function . The knowledge storing capacity of the intelligent neuron is analyzed. It is proved that the intelligent neuron has greater knowledge storing capacity than normal neuron and can strengthen the information processing capability of neural network. In the paper, traffic flow data of DALIAN city is predicted by applying such neural network. Experimental results show that intelligent neural network has higher astringency rate and precision, and has greater advantages than BP network.Promoted by the research work in ITS, dynamic traffic assignment (DTA) theory became a focus research area. Genetic algorithm (GA), as an excellent global optimization algorithm, has been applied extensively. Based on above theoretical bases, in this paper an optimal control model of Master/Slave parallel genetic algorithm to solve DTA problem is proposed. Based on the Drawing cluster with distributed storage and message passing system, this algorithm is implemented in the Master/Slave mode of the PVM parallel platform. An example transportation network is applied to test the model. The experimental results show excellent system operation indices and parallel efficiency, which prove the validity of the model.The application of grid computing on scientific research and engineering problem is support by aggregate distributed high performance compute , large database, scientific research equipment and other substantial computational resources to form a supercomputer. Base on exist hardware and network equipment the grid computing experimental environment is constituted in this paper. The parallel Dijkstra algorithm to solve the short path problem is tested in the experimental environment. Base on exist hardware and network equipment of DUT and DALIAN traffic control system, the DUT computing grid model and traffic control grid system are designed. The implement of traffic flow prediction on grid environment is discussed base on this model.更多还原