【作者】 李存军；

【导师】 杨儒贵；

【作者基本信息】 西南交通大学 ， 交通信息工程及控制， 2004， 博士

【摘要】 随着经济的发展和社会的进步，交通问题日益突出。作为有效解决现有道路交通问题的途径，发达国家正在竞相研究智能运输系统(ITS)。ITS将成为21世纪现代化地面交通运输体系的发展方向，是交通运输进入信息时代的重要标志。以交通流量为重要内容的道路交通信息是所有ITS项目不可缺少的基础，准确地预测交通流量是ITS的关键所在，也是交通流诱导研究的重要环节。交通流量是一个时变的非线性系统，其内部变量、输入变量众多，结构复杂。采用部分因素和指标或单个的模型仅能体现系统的局部，多个变量的科学综合、多种数据或多个模型的有效融合能够显著提高预测精度和模拟效果。基于此，本文提出了基于集成神经网络的城市道路交通流量的融合预测模型。 该融合预测模型和传统的预测方法相比，既不是单一的数据预测方法的运用，也不是对单一预测数据的使用，更不是对这些方法和数据的简单组合，而是包括从拓展数据源到选择有效的预测方法，再到方法和数据的融合。该模型一方面提高了预测的准确性，另一方面提高了预测的鲁棒性。 本文以城市道路交叉口为例，对基于集成神经网络的融合交通流量预测的模型、方法和具体实现途径进行了研究，主要包括以下方面： 1)根据融合预测的需要分析了交通流量的分布特性：周期性、连续性和相关性。这些特性反映了交通流量的决定因素和影响因素； 2)建立了路口交通流量融合预测的多源数据模型，扩展了预测的数据基础，为预测的准确性和鲁棒性准备了条件； 3)提出了根据流量周期性的灰色神经网络预测方法； 4)应用小波分析理论对交通流量进行多分辨率分析，提出了根据流量连续性的小波神经网络预测方法； 5)提出了对各路口流量进行实时相关性分析的方法，和一定日期内相应时间段路口流量相关性的模糊综合评价方法，提出了利用路网中相关路口的数据预测流量的神经网络方法； 6)提出了融合预测模型中局部预测有效性检验的聚类分析方法以西南交通大学博士研究生学位论文第H页及实现最终融合预测的模糊神经网络模型。 本文在仿真生成的流量数据的基础上，对上述模型和方法的有效性进行了验证。 关键字:交通工程，神经网络，数据融合，交通流量预测，灰色理沦，小波分析，相关性分析，模糊评价，有效性检验，聚类分析，模糊系统更多还原

【Abstract】 With rapid development of economy and society, the problem caused by traffic is becoming more and more serious. Intelligent Transportation System (ITS), which is widely studied in developed countries, is thought of a potent way to the ground traffic problem today. ITS which is an important symbol of ground transportation system will come in the 21st century. Ground traffic-information, which mainly consists of traffic-flow, is an indispensable element of ITS. It is a key of ITS to predict traffic-flow exactly which is an important step in the research of traffic-flow guidance system. Traffic-flow is a time-varying nonlinear system that is comparatively complex in structure and internal variable. A lot of variables should been input. It only embodies any aspect of system to apply any factor or single model. lt notably increases prediction precision and simulation performance by combining multiple variables, fusing data and models. Accordingly, this paper proposes a fusion-prediction model of traffic-flow in urban road-intersection based on integrated ANN (Artificial Neural Network).Compared with traditional method, this model is not meant to apply single prediction technique or one-sided message, or to fit these together simplly, but meant to widen source of data, to optimize prediction way, and to fuse data and methods soundly. On the one hand, it can evidently increase prediction accuracy; on the other hand, robustness of system is markedly enhanced.With an example of urban road-intersection, how to build the data-fusion prediction model of traffic-flow volume, and how to effectively apply these methods stated in the model, is detailly studied in this paper as follows:1) To meet the need of data-fusion prediction, the features (periodicity, continuity and correlation) of traffic-flow volume are analyzed specislly, which reflect the decisive and influencing factors of traffic-flow.2) It is aimed at enhancing accuracy and robustness that multi-source data model of traffic-flow in road-intersection is built and accordingly data base of prediction is expanded.3) Prediction model of GNN (Gray Neural Network) is put forward according to the periodicity of traffic-flow.4) Prediction method of WNN (Wavelet Neural Network) is put forward according to the continuity of traffic-flow with Wavelet MRA in analysis of traffic-flow volume.5) Neural network method for traffic-flow volume prediction by corresponding road-intersection in road-net is put forward after analyzing correlation in real-time segment of traffic-flow volume between different road-intersections, estimating correlation of traffic-flow volume in road-intersections in corresponding time-segment with fuzzy comprehensive evaluation method.6) Final data-fusion prediction model on FNN (Fuzzy Ncural Nctwork) is stated also with the clustering analysis method of verifying validity in local predictionsThese methods and algorithm is tested by simulation.更多还原