【作者】 任黎立；

【导师】 彭其渊；

【作者基本信息】 西南交通大学 ， 交通运输规划与管理， 2005， 硕士

【摘要】 随着交通量的增加,高速公路上交通拥挤和阻塞日趋增多,降低了高速公路的运行效率和安全性,加剧了环境污染和能源消耗。入口匝道控制通过控制进入高速公路的车辆数目使高速公路交通流运行在最佳状态,可以有效提高高速公路运行效率,避免拥挤和阻塞,如何对入口匝道实施有效且鲁棒性好的控制是值得研究的课题。鉴于人工神经网络具有良好的非线性特性、学习能力、自适应能力和抗干扰能力,因此本文将人工神经网络技术引入到高速公路入口匝道控制中,对其进行研究。 本文首先介绍了入口匝道控制和神经网络理论的背景知识。选用结构简单,计算方便,具有多输入、多输出特性的BP神经网络构造控制器。并在对BP神经网络学习算法进行推导的基础上,详细讨论了BP神经网络控制器的设计过程,包括输入/输出层的设计、输入/输出数据的处理、隐含层单元数的选取、初始权值的设定、学习速率的选取等方面,并针对高速公路入口匝道控制的具体的应用提出了设计原则。 其次,论文设计了两种用于入口匝道控制的神经网络控制器——直接型和积分型。由于学习样本收集困难,所以设计时将控制器作为控制系统的一部分,引入了BP改进算法训练神经网络控制器。直接型神经网络控制器设计为采用前六个时间点的状态信息作为输入量直接产生控制量,而积分型神经网络控制器则是将期望值与输出值之间偏差的积分构成控制量对被控对象进行控制。本文对控制器的控制效果进行了仿真,并将控制结果与ALINEA、需求容量控制和不加控制下的情况进行比较,结果表明:神经网络控制可提高控制的鲁棒性,更好地稳定主线交通密度。 最后,考虑入口匝道排队长度对控制效果的影响,对上述的神经网络控制器进行改进,提出了两种可抑制排队长度增长的方法,并对控制效果进行了仿真,将其与ALINEA控制以及不加控制的情况进行了比较,结果表明:提出的两种方法均对减少匝道内排队长度十分有效,可以大大减轻对临近道路交通干扰。 研究表明将BP神经网络技术引入到高速公路的单个入口匝道控制中,对于避免路段拥挤、道路密度波动、抑制匝道排队长度增长等具有很好的效果。更多还原

【Abstract】 Nowadays recurrent and nonrecurrent congestion on freeways is increasing, which causes delays, environmental pollution and traffic unsafety. Ramp metering is one of the most efficient and wide-used means to alleviate this phenomenon. The efficient and robust control of the on-ramp is a hot research topic with great practical value. Due to the traits of nonlinear, capacity of study, self adapting and anti-interference, neural network is suitable for the control of ramp metering.Firstly, the background of ramp control and neural networks was introduced. Back-propagation (BP) neural network has many merits, such as simple structure, easy to calculate and multi-input/multi output, which makes it to be greatly applied in control. Therefore, in this thesis the algorithm of BP was introduced and the design of BP control was discussed in detail, including the design of input/output (I/O) layer, the processing of the data of I/O, the selection of the number of hide layer and learning rate etc.Secondly, two neural networks controllers for the ramp control are developed, which are the direct controller and the PI controller. The controller is considered as part of the control system to overcome the difficult of sample collecting. The neural networks controller is trained by adjusted BP algorithm. Preceding traffic conditions is put into the direct networks controller, whose outputs are ramp metering rate. The difference between the output and the reference is put into the PI neural networks controller. The control effect is simulated with random test data and compared with the ALINEA, demand-capacity control and without control. It shows that the neural networks control has robustness, and the mainline traffic density is more stable.Finally, the neural networks controller is improved when the queue of vehicles on the ramp is considered. Upon which two control techniques are brought out to reduce the queue length. The control effect is simulated and compared with ALINEA control and without control. It’s shown that the methods developed are helpful to reduce the queue of vehicles and the interference with surface street traffic.In a word, BP neural networks controllers in the freeway ramp control can avoid freeway jam greatly, make mainline traffic condition stable and diminish the queue of vehicles on the ramp.更多还原