小波神经网络算法及其在交通流宏观动态特性中的应用

发布时间：2018-03-29 23:13

本文选题：交通流　切入点：宏观特性　出处：《重庆大学》2014年硕士论文

【摘要】：随着科学技术的不断进步，国家的工业化水平得到提升，大大促进了经济的发展，汽车行业在这个过程中获得机遇。然而，这种发展给社会带来了诸多问题，如能源过度消耗、大气污染、交通拥堵等，其中交通拥堵带来的损失不可估量。研究交通拥堵问题，主要是针对道路交通流的整体运行情况进行研究，交通流数据中宏观动态特性主要是指交通流宏观参数即交通流量和速度的变化，，它们易于获得而且能较好表征交通流的状态。交通流宏观参数具有很强的随机性、非线性等特点，但是在短期内又具有一定的准周期性。分析交通流动态宏观特性的典型代表方法分为两种：有数学模型的ARIMA和无数学模型的神经网络。传统的分析方法存在很多的不足，一些非线性、非平稳数据分析需要多种综合的方法，这些系统模型很难用确切的数学模型表达出来，而神经网络不需要建立精确数学模型，并且能够达到较好的效果。本文采用小波神经网络对历史交通流数据进行分析，主要研究工作如下： ①首先对神经网络原理、小波知识进行介绍，对小波神经网络进行研究，然后建立小波神经网络模型；对采集到的GPS交通流数据进行预处理，包括数据异常修复、错误替换等操作，通过测试数据对小波神经网络模型进行仿真实验。 ②针对误差较大的问题，首先采用增加动量项、动态学习系数等方法对小波神经网络算法进行改进；然后针对小波神经网络存在的缺陷进行分析，采用遗传算法对小波神经网络进行优化，针对遗传算法中的交叉率和变异率采用改进的动态自适应算法，并对改进的算法进行验证。 ③采用改进后的概率统计算法，实现交通流数据的电子地图匹配系统。通过遗传算法优化后的小波神经网络，对交通流宏观参数历史数据进行仿真，分析其变化规律；然后介绍城市道路交通状态，分析宏观参数的变化对道路交通状态的影响，最后得出结论并对今后工作进行展望。
[Abstract]:With the continuous progress of science and technology, the industrialization level of the country has been improved, greatly promoted the economic development, the automobile industry obtains the opportunity in this process.However, this development has brought many problems to the society, such as excessive consumption of energy, air pollution, traffic congestion and so on, among which the loss caused by traffic congestion is incalculable.The study of traffic congestion is mainly aimed at the overall operation of road traffic flow. The macroscopic dynamic characteristics of traffic flow data mainly refer to the changes of traffic flow macroscopic parameters, namely, traffic flow and speed.They are easily available and can better represent the state of traffic flow.The macroscopic parameters of traffic flow are characterized by strong randomness and nonlinearity, but they are quasi-periodic in the short term.There are two typical methods to analyze the macroscopic characteristics of traffic flow: ARIMA with mathematical model and neural network without mathematical model.There are many shortcomings in the traditional analysis methods, some nonlinear and non-stationary data analysis need a variety of integrated methods, these system models are difficult to express in exact mathematical models, but neural networks do not need to establish accurate mathematical models.And can achieve better results.In this paper, wavelet neural network is used to analyze the historical traffic flow data. The main research work is as follows:Firstly, the principle of neural network and wavelet knowledge are introduced, then the wavelet neural network model is established, and the collected GPS traffic flow data are preprocessed, including abnormal data repair.The wavelet neural network model is simulated by test data.(2) aiming at the problem of large error, the wavelet neural network algorithm is improved by increasing momentum term and dynamic learning coefficient, and then the defects of wavelet neural network are analyzed.The genetic algorithm is used to optimize the wavelet neural network, and the improved dynamic adaptive algorithm is adopted for the crossover rate and mutation rate of the genetic algorithm, and the improved algorithm is verified.3 using the improved probability and statistics algorithm to realize the electronic map matching system of traffic flow data.Based on the wavelet neural network optimized by genetic algorithm, the historical data of macroscopic parameters of traffic flow are simulated, and the changing law of traffic parameters is analyzed, and then the traffic state of urban road is introduced, and the influence of the change of macroscopic parameters on the state of road traffic is analyzed.Finally, the conclusion is drawn and the future work is prospected.
【学位授予单位】：重庆大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：U491.112

【参考文献】