基于神经网络的城市道路交通量短时预测研究

发布时间：2019-01-16 01:12

【摘要】：智能交通系统(ITS)能成功应用于城市道路网的关键技术之一,是城市道路交通量短时预测。满足系统精度要求的城市道路短时交通量,是支撑交通管理系统、出行者信息系统等智能交通系统各子系统的基础。准确的交通量短时预测数据,可应用于道路交叉口信号灯实时配时,减少红灯延误并增强交叉口通行能力;可增强交通流可视化应用的时效性,改善城市车辆在路网运行的安全状况。城市道路交通量短时预测有着十分广阔的应用前景,但交通量短时预测具有随机性与非线性等特征,使其建立数学模型较为困难。在认真研究了交通量短时预测的各项特征之后,本文认为人工神经网络的诸多优点均可与其随机性、非线性等特征相契合。将预测模型与神经网络二者理论综合考量与融合后,建立以人工神经网络预测模型为基础的针对城市道路交通量的短时预测模型。利用MATLAB软件实现预测模型之后,选取实例验证其准确性,具体研究内容如下:首先,对交通量短时预测理论进行研究,并利用短时交通量的时空特征对短时预测模型进行分类,在此基础之上探究模型建立的原则与制定相应的模型评价指标。其次,在对人工神经网络的相关理论进行研究后,提出将BP神经网络与交通量短时预测相结合的预测方案,建立基于BP神经网络的交通量短时预测模型。建立这种短时预测模型的关键在于BP神经网络模型拓扑结构的建立与相关参数的确立,在讨论模型网络层数选取,神经元数目确定,预测数据处理等模型建立关键步骤的同时,也对BP神经网络的局限性进行分析讨论,针对其局限性提出优化改进的措施。再次,针对前文提出的优化BP神经网络权值与阈值来避免网络模型陷入局部最小值这一思路,选用具有自适应全局优化搜索算法的遗传算法建立遗传-BP神经网络交通量短时预测模型。利用遗传算法优化网络模型的权值与阈值之后,再进行网络模型的训练及仿真预测,以提高模型的预测准确性。最后,以银川市市区交叉口实测的短期交通量数据为样本,根据短时预测模型的分类确立两种不同的输入方案;根据前文对BP神经网络以及利用遗传算法对网络的优化分析,运用MATLAB2009a神经网络工具箱建立基于传统BP神经网络模型与基于遗传算法优化传统BP神经网络模型权值与阈值的两类交通量短时预测模型。利用两种方案与不同模型相结合对研究交叉口进口道交通量进行预测,并对预测结果进行分析与对比。结果表明,BP神经网络模型预测结果基本满足应用要求;优化后的遗传-BP神经网络模型避免了BP神经网络的缺陷,提高了预测精度,更具利用价值。
[Abstract]:Intelligent Transportation system (ITS) is one of the key technologies which can be successfully applied to urban road network, and it is the short-term prediction of urban road traffic volume. The short-time traffic volume of urban road, which meets the requirement of system precision, is the basis of supporting the subsystems of intelligent transportation system, such as traffic management system, traveler information system and so on. Accurate short-term traffic volume prediction data can be applied to real-time traffic signal timing at road intersections to reduce red light delay and enhance traffic capacity at intersections. It can enhance the timeliness of traffic visualization and improve the safety of urban vehicles in road network operation. Short-term prediction of urban road traffic volume has a very broad application prospect, but short-term traffic volume prediction has the characteristics of randomness and nonlinearity, which makes it difficult to establish mathematical model. After studying the characteristics of traffic volume prediction in a short time, this paper holds that the advantages of artificial neural network are consistent with its randomness and nonlinearity. After synthesizing the theory of forecasting model and neural network, a short-term forecasting model for urban road traffic volume is established based on artificial neural network prediction model. After using MATLAB software to realize the prediction model, an example is selected to verify its accuracy. The specific research contents are as follows: firstly, the theory of short-term traffic volume prediction is studied, and the short-time forecasting model is classified by using the space-time characteristics of short-term traffic volume. On the basis of this, the principle of model establishment and the corresponding evaluation index of model are explored. Secondly, after studying the related theory of artificial neural network, a short-term traffic volume prediction model based on BP neural network is established by combining BP neural network with short-term traffic volume prediction. The key to establish this short-term prediction model lies in the establishment of topological structure of BP neural network model and the establishment of related parameters. While discussing the key steps of establishing the model, such as the selection of network layer number, the number of neurons, the prediction data processing and so on, the key steps of establishing the model are discussed. The limitation of BP neural network is analyzed and discussed, and the measures to optimize and improve it are put forward. Thirdly, to optimize the weights and thresholds of BP neural network to avoid the network model falling into the local minimum, The genetic algorithm with adaptive global optimization search algorithm is used to establish the short-term traffic prediction model of genetic BP neural network. After the weights and thresholds of the network model are optimized by genetic algorithm, the training and simulation prediction of the network model are carried out to improve the prediction accuracy of the network model. Finally, based on the short-term traffic volume data measured at the intersection in Yinchuan city, two different input schemes are established according to the classification of short-term prediction model. According to the BP neural network and the genetic algorithm to optimize the analysis of the network, Using the MATLAB2009a neural network toolbox, two kinds of short-term traffic volume prediction models based on traditional BP neural network model and genetic algorithm optimized traditional BP neural network model weight and threshold are established. Combining two schemes with different models, the traffic volume of the intersections is forecasted, and the results are analyzed and compared. The results show that the prediction results of BP neural network model basically meet the requirements of application, and the optimized genetic BP neural network model avoids the defects of BP neural network, improves the prediction accuracy and has more value in utilization.
【学位授予单位】：兰州交通大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：U491.14;TP183

【相似文献】