Research On Methods Of Traffic Flow Predicting Of The Urban Road Network Based On The Multi Cross-section Information

With the development of social economy, the gross quantity of urban cars climbs rapidly, bringing great pressure for the urban road transport system. Traffic problems, such as, traffic accidents, traffic jam, environment, energy and so on, are getting more and more series. All of these greatly reduce the conveniences of people traveling. The Intelligent Transport System (ITS) is a effective method to solve urban road traffic problems. One of the basic works is traffic flow prediction, which provides theoretical and data support for ITS, gives real-time effective traffic information for travelers and induces their travel behaviors. So it is great helpful to make full use of transportation infrastructures and solve or relieve traffic problems. In recent years, it is a hot spot to deeply analyze and get the changing law, improve the real time, reliability and self-adaptability of traffic flow prediction.This thesis takes urban road transport network as study objects, explores the spatial-temporal characteristics of the traffic flow and spatial-temporal prediction theory deeply. Its main study contents and results are given as follows.The urban road traffic flow changes over time, which is affected by space factors. Based on the analysis of the traffic flow spatial variations, some influence factors on the traffic flow spatial interaction, such as distance, traffic states and so on, are analyzed emphatically. Meanwhile, delay characteristics of traffic flow spatial interaction is analyzed in detail referring to the urban geography theories. A computing formula of traffic flow spatial correlation based on delay characteristics is given. The validity of this formula is verified through quantitative analysis.After comparing some clustering methods, a new clustering algorithm of road cross-sections based on the spatial correlation of road traffic flow is givenSpatial-temporal prediction considers spatial-temporal variation characteristics of the traffic flow, which makes multiple cross-sections as its study objects. So, if all of cross-sections are analyzed and predicted, it will greatly increase the computational complexity. After comparing some clustering methods, a new clustering algorithm of road cross-sections based on the spatial correlation of the road traffic flow is given, which introduces average correlation, principal component analysis and k-means. It doesn’t need to initialize clustering parameters, such as clustering number, clustering centers and so on, so it is highly flexible and self-adaptive.The thesis explores traffic flow prediction of single and multiple cross-sections. Referring to the idea of process neural network prediction model, the Online Adaptive Hybrid Spatial-temporal Short-term prediction model (OAHST model) is proposed. OAHST model is a single-input-multiple-output model, which divides the road cross-sections into main cross-section and assisting cross-sections. It first predicts the traffic flow variation of each cross-section, and finally the predicting result of the main cross-section is revised according to the correlativity between cross-section traffic flow.Under the condition rich initial traffic flow samples, an online self-adaptive RBF network prediction algorithm is constructed. In the process of establishing the neural network, Determining the network structure is a difficult point, that is basic information of hidden nodes of neural network. Appropriate number of hidden nodes can make computational complexity model lower, improve the training convergence speed. After deeply exploring the sequential learning algorithm of the neural network and Radial Basis Function neural network (RBFNN), the Two-Stage Mixed self-adaptive Learning Algorithm of the RBFNN (TSMALA) is proposed, which dynamically determines the network structure. In TSMALA, RBFNN dynamically adds and deletes hidden nodes according to the importance of hidden nodes. This can make the prediction model respond to external changes quickly and efficiently.Under the condition of less initial samples, the online self-adaptive least squares support vector machine prediction algorithm is established. Least squares support vector machine (LS-SVM) is the optimization of support vector machine (SVM). It makes all samples as support vectors, which lead to increases the computational complexity of the model, make the results of the model lack sparsity and slow down its testing speed. Based on these and exploring support vectors, the online self-adaptive LS-SVM prediction model (OALS-SVM) is proposed, which dynamically adds, deletes support vectors, strictly limits the number of support vectors, and makes the prediction model respond to external changes quickly and efficiently.According to the real requirements, urban road traffic flow guiding system is developed as the case study based on the researching results. At last, the main algorithms and several operation interfaces are presented.The thesis deeply explores spatial-temporal variation characteristics of traffic flow, referring to the idea of the process neural network prediction model, a novel spatial-temporal hybrid prediction model (OAHST model) is proposed. OAHST model considers the input dimensions individually, establishes the moving data window of traffic flow, and dynamically determines the inputs of the prediction model according to the input samples. Based on different size of the Initial samples, two prediction algorithms of the single cross-section is proposed, which are based on LS-SVM and RBFNN respectively. These greatly improve the self-adaptability and flexibility of the model. By the way of combining theory with practice, the aims of the research are achieved. It is a useful exploration process for solving urban road traffic problems and developing the ITS.