Research On The Key Technologies Of Traffic State Evaluation And Forecasting For Urban Expressway

With the constant improvement of urban road traffic networks,urban expressway plays an irreplaceable role in urban traffic system. The operational quality of expressway not only affects the efficiency of the whole urban road network, but also affects the normal urban function exertion. However, with the rapid increases of traffic demand,urban expressway inevitably appears the phenomenon of traffic congestion. The congestion propagation of expressway is more serious, which greatly limits the function of the expressway such as comfort, convenience and safety. In addition, being the highest rank of urban road traffic network system, urban expressway equipments with perfect traffic detectors, which can obtain dynamic traffic data with different accuracy, breadth and content. It provides enough data for monitoring the traffic state of urban expressway dynamically. Therefore, the keys to expressway traffic management and control are to make full use of traffic data resources, to evaluate expressway traffic state accurately and forecast the development trend of traffic state.In this paper, based on the time and space characteristics of urban expressway traffic flow data,an intensive study has been conducted on the key technologies of traffic state evaluation and forecasting for urban expressway, which provides the theoretical basis and technical support for traffic management and control of urban expressway.The main research achievements are as follows.(1) The traffic flow missing data completion method for urban expresswayIn view of the limitation of dimension for vector and matrix, the multi-mode correlation information of expressway traffic flow data can not be fully utilized. In this paper, the concept of tensor was introduced to recover the traffic flow missing data. On the basis of analyzing the spatial and temporal correlation of traffic flow time series data, the hierarchical Tucker tensor decomposition model was proposed to recover the traffic flow missing data, and the optimization algorithm based on Riemann manifold was used to solve the problem. Finally, validation analysis was carried out using traffic flow data measured from expressway.(2) The multi-scale traffic data fusion method for urban expresswayIn regard to the problem that most of the current traffic data fusion methods focus on feature-level fusion or decision-level fusion. Based on multi-detectors sampling analysis, asynchronous sampling was taken as the research object. Then, a multi-scale data fusion algorithm for urban expressway based on wavelet transform and Kalman filtering was proposed. The effectiveness of the algorithm was verified by means of simulation data and measured data from expressway. The study results could ensure the quality of traffic data from the source, which provides strong data support for the follow-up study.(3) The automatic traffic incident detection method for urban expresswayOn the basis of analyzing the change regulation of expressway traffic parameters at the time of traffic incident, an initial set of traffic incident detection which consists of 12 variables was constructed. The importance of variables measured by random forest model was used to filter out the key variables which are more sensitive to traffic incident. Then, the relevance vector machine model based on particle swarm optimization was constructed. Aiming at the problem of uneven distribution of training sample data set, the SMOTE method was used to reconstruct sample data set. Finally, validation analysis was carried out using loop data measured from expressway.(4) The automatic congestion identification method for urban expresswayOn the basis of illustrating the classification of traffic state and its measurement standards for expressway, the traffic flow, speed, occupancy, the ratio between occupancy and traffic flow, the ratio between occupancy and speed were selected as traffic state characteristic variable. A traffic state identification method based on projection pursuit dynamic cluster model was proposed. First, dynamic cluster method was used to constructed the projection index function. Then, the shuffled frog leaping algorithm was used to obtain optimum projective direction. Finally, validation analysis was carried out using both simulated data and measured data.(5) Traffic state forecasting method for urban expresswayThis part includes two aspects of content, one is the short-term traffic parameters forcasting method based on cointegration theory, the other one is short-term traffic flow local forecasting method based on chaos theory. In the aspect of cointegration theory, after the stationarity test, the determination of lag order, cointegration test and parameters estimation, short–term traffic flow prediction methods based on traffic flow-speed-occupancy vector error correction model and speed-occupancy vector error correction model were constructed. Then, the stationarity of constructed model was tested and the pulse response analysis was conducted. In the aspect of chaos theory, the C-C method was used to reconstruct phase-space, and the number of neighboring points was determined by use of Hannan-Quinn criteria. Then, a short–term traffic flow prediction method based on combined kernel function relevance vector machine model was constructed. Finally, validation analysis was carried out using loop data measured from expressway.