【作者】 杜长海；

【导师】 黄席樾；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 随着社会经济的高速发展与城市化进程的不断加快,城市人口和机动车辆日益增加,城市道路交通拥堵已经成为世界大中城市普遍存在的现象,由此带来的交通事故、能源浪费以及环境污染等问题,不仅严重地制约着城市和社会经济的可持续发展,同时也严重地影响着城市居民的生活质量。因此,运用智能交通系统来解决日益严重的城市交通问题,已经成为了交通工程未来发展的重要方向。本学位论文研究的交通诱导系统是智能交通系统中的重要子系统之一,它能够有效地引导车辆在路网中运行,减少车辆在道路上的行驶时间,并最终实现交通量在整个路网中均匀分配。将自然科学的最新研究成果和工程技术的最新方法引入城市交通诱导系统,不仅有利于提高交通系统的运行效率,而且关系到土地资源与能源的合理利用、环境污染与噪声的改善,这对满足社会需求、推动国家和社会的进步以及学科的发展,无疑都具有十分重要的意义。城市交通系统集成了人、车、路和环境等各种因素,具有高度的复杂性、时变性和不确定性。精确的数学模型和方法难以有效解决复杂的现代城市交通问题,而计算智能是一种仿生计算方法,它从生物底层对智能行为进行模拟和研究,拓展了传统的计算模式,不需要建立问题本身的精确模型,具有智能性、并行性、自适应性等优点,为复杂问题的求解提供了卓有成效的解决途径。因此,本学位论文依托重庆市科学技术委员会自然科学基金计划项目——智能交通系统重点项目“城市交通路网拥堵动态预警与疏导决策技术研究”(项目编号:CSTC,2006BA6016),在总结该领域现有研究成果的基础上,引入新兴起的计算智能理论,对交通诱导系统研究中应解决的若干关键理论问题进行了较为全面深入的研究,试图解决目前城市交通诱导领域存在的一些疑难问题。本文主要的创新性工作如下:①针对支持向量回归机(SVR)的拟合精度和泛化能力取决于相关参数的选取问题,提出了基于灾变FS算法的SVR参数选择方法,并应用于交通流预测的研究。通过提出基于尖点模型的灾变策略来改进FS算法的个体迭代位置选择机制,以降低设置搜索半径的依赖性,从而扩大搜索空间,提高全局搜索能力。对实测交通流量进行滚动预测实验,结果表明该方法优化SVR参数是有效、可行的,与经验估计法相比,得到的SVR模型具有更好的泛化性能和预测精度。②针对基本蚁群算法存在容易陷入局部最优解出现早熟停滞的缺点,提出了基于混沌选择策略的蚁群算法,并应用于城市交通路径寻优的研究。在基本蚁群算法的概率转移中引入混沌扰动的策略,以使解易于跳出局部极值区间,加快收敛速度。以重庆市渝中半岛的路网为实例计算以行程时间为目标的最优路径,结果表明该算法获得了较好的效果,与基本蚁群算法相比,提高了全局搜索能力。③针对牛顿法初始值要求严格、易产生局部收敛并含有矩阵求逆的不足和粒子群算法存在收敛速度慢和局部最优的问题,引入粒子间相对位置改进基于抗体浓度的概率选择公式,提出了一种带免疫机理的粒子群算法,并将其用于由路段流量反推OD矩阵的极大熵模型求解研究。粒子不仅根据个体极值和全局极值更新速度和位置,而且按一定概率以轮盘赌法选择某个粒子进行学习,以保持种群多样性,降低了算法过早收敛于局部最优解的几率。以重庆市某交叉路口为实例进行实验,粒子群算法求解成功率高于牛顿法,表明粒子群算法推算OD矩阵是一种行之有效的方法;而且,改进的粒子群算法比基本粒子群算法和基本遗传算法具有更好的全局寻优能力。④针对城市交通状态的不确定性和模糊C均值聚类(FCM)算法存在的初值敏感性和局部搜索性问题,提出了基于混合蛙跳算法(SFLA)的模糊C均值聚类算法(SFLA-FCM),并应用于城市交通状态识别研究。SFLA-FCM使用SFLA的优化过程代替FCM的基于梯度下降的迭代过程,有效地避免了FCM对初值敏感及容易陷入局部极小的缺陷。实验结果表明,与FCM相比,SFLA-FCM在收敛速度和精度上均有所提高,结果更为合理、稳定;而且,能够有效地对交通状态进行识别,为交通状态实时识别提供了一个新的研究思路。总之,本文对计算智能的理论和方法在城市交通诱导系统领域的应用进行了较为全面深入的分析研究,对进一步解决城市交通问题具有重要的意义和广阔的应用前景。更多还原

【Abstract】 With the increasing development of social economy and urbanization, urban population and vehicles increase rapidly. Traffic congestion has become a prevalent problem for metropolis all over the world. Traffic accident, energy wasting, and air pollution resulted from traffic congestion not only seriously restrict the sustainable development of social economy, but also severely influence the urban living environment. So adopting intelligent transportation system (ITS) technology to solve these problems has become an important direction in the future transportation engineering development.Traffic guidance system is an important subsystem of ITS, which can lead vehicles to move in road network effectively, reduce driving time, and finally realize that the traffic volume distributes equably in the whole road network. Applying the newest research of science and technology to urban traffic guidance system can not only improve transportation safety, production efficiency and revenues, but also connect with land resource and energy exploitation, environment improvement, which is of the most important significance to meet social demand, accelerating the progress of nation and society, and driving subject development.Urban road traffic system, which integrates human, vehicles, roads, environment and other complex factors, is of high complexity, time-dependence and randomicity. It is difficult for traditional control methods based on precise mathematical models to solve complex modern urban traffic problems. And computational intelligence (CI) is a computing methodology from nature, which simulates and researches the intelligent behavior from the lowest level of the creature. CI develops the traditional style of computation without establishing complicated mathematical models, which has many advantages, such as intelligence, parallel processing and self-adaptive ability. Thus it provides a fruitful approach to solve complex problems.Stemming from the key project——“Research on Dynamic Congestion Warning and Evanesce Decision-making for Urban Traffic Network”, funded by Natural Science Foundation Project of CQ CSTC under Grant 2006BA6016,on the basis of reviewing the existing outcome in this field, this dissertation introduces computing intelligence theory to make a comprehensive and deep research on several important problems from urban traffic guidance system. The main original points of this paper lie below:①Regression accuracy and generalization performance of support vector regression (SVR) models depend on proper setting of its parameters. An optimal selection approach of SVR parameters is proposed based on catastrophic FS algorithm and applied on traffic flow forecasting. The decision mechanism of individual initial position is improved through introducing cusp-catastrophe strategy to reduce reliability on search radius, and to extend the area of feasible solutions and enhance the global search ability. Through a rolling forecasting simulation experiment on real traffic volumes, the experimental results show that the proposed method is feasible and effective for the optimal selection of SVR parameters, and has better generalized performance and prediction accuracy than rule of thumb.②Due to the disadvantage of local optimum of basic ant colony algorithm, chaotic selection strategy are proposed to improve ant colony algorithm, and applied in optimal route of urban road network. Chaos perturbation is used to improve selection strategy to avoid precocity and stagnation. The road network of Chongqing Yuzhong Peninsula is taken as an example to calculate the optimal route based on the least travel time, and the experimental results show that this algorithm has much higher capacity of global optimization than basic ant colony algorithm and it is feasible and effective for optimal route choice.③Due to the disadvantage of slow convergence and local optimum of particle swarm algorithm, introducing relative distances among particles to improve probability selection formula, an improved particle swarm optimization with immune mechanism is proposed. Particles update their velocity and position not only by individual and global optima, but also by individual optima of a specific particle selected by roulette method according to certain probability, to keep the variety of the population and avoid precocity and stagnation. This algorithm is applied to solve the maximum entropy model, estimating OD matrix from traffic link flows. Through a test on a specific crossroad in Chongqing City, the experimental results show that particle swarm algorithm is feasible and effective for OD matrix estimation, overcomes the shortcoming of Newton’s method that strictly depends on initial values, and the particle swarm algorithm has much higher capacity of optimization than basic particle swarm algorithm and basic genetic algorithm.④Due to uncertainty of urban traffic state, and initialization sensitivity and local search of fuzzy c means (FCM), a new FCM algorithm (SFLA-FCM) based on shuffled frog leaping algorithm (SFLA) is proposed and applied on study about urban traffic state. SFLA is a new recta-heuristic population evolutionary algorithm and it has fast calculation speed and excellent global search capability. SFLA-FCM uses SFLA to replace the iteration process of FCM based on the gradient descent and avoids the disadvantages of local optimality and initialization dependence. The experimental results show that the proposed method is more accurate and efficient than FCM and it is feasible and effective for traffic state identification.In short, an in-depth study for urban traffic guidance system is carried out comprehensively by computing intelligence theory and method, which has great significance of solving urban traffic prolems furtherly.更多还原