Urban Road Traffic State Evaluation And Prediction-a New Scheme With Applications

The complete acquisition, accurate/real-time evaluation and prediction of urban road traffic state, is the foundation to grasp accurate behaviours of urban road traffic system, make scientific decisions of traffic management, and make full use of traffic facilities. However, how to define the traffic state, how to quantitatively describe the traffic state, and how to evaluate and forecast the traffic state in multi-hierarchy and multi-granule for traffic participants, have been the difficulty and emphasis in the field of transportation research. Moreover, the study of those problems is of great theoretical and practical significance.This dissertation, on the background of actual traffic management, elaborates the traffic state’s concept system, deeply studies the traffic state’s concept system, the methods of traffic state acquisition, real-time evaluation and prediction, and then verifies their validity by practical application, which formed a relatively complete methodology system of traffic state evaluation and prediction; The main innovative work of this dissertation is specifically summarized as follows:1) A concept system of traffic state and its application framework are formed.Based on the analysis of urban road traffic system, the traffic state’s definition, elements, attributes, and categories are illuminated profoundly, which compose a more complete concept system of urban road traffic state. By analyzing the traffic state’s formation and the demands of actual traffic management, an application framework based on the evaluation and prediction method of traffic state is designed to support the traffic management decision-making, which provides a foundation to further research.2) The solutions of two typical data "missing" problem are proposed to acquire the complete, accurate traffic state.A soft-sensing method based on pre-selection space-time model is proposed to solve the detection "missing" data. With the existing detection data, the model firstly makes use of the spatial and temporal correlation among the traffic data sequences to construct the space-time model; and then reduces its estimation parameters through the pre-selection strategy, which can improve its effectiveness and efficiency. To solve the accuracy "missing" data, the concept of source-credibility is introduced to describe the degree of the detection data approaching the true value; and then two fusion models based on source-credibility and knowledge, source-credibility and approximate reasoning are proposed respectively according to the division of source-credibility. The two fusion models not only consider the time-dependent source-credibility, but also integrate the experts’experience and reasoning respectively, which made the regularity of the traffic data and the knowledge of experts unified.3) Two real-time traffic state evaluation methods coinciding with traffic participants’cognition are established to satisfy the demands of actual traffic management.The connotation of the traffic state evaluation, which can be considered as a logical process of judging the traffic state’s category by traffic participants, is analyzed firstly; and then the way to evaluate the traffic state from the point of traffic managers is clarified; following that, the evaluation indexes derived from the real-time traffic data are designed, and the real-time traffic state evaluation method based on subjective and objective integrated experiments for road section, the real-time evaluation method based on fuzzy clustering and fuzzy comprehensive evaluation model for road network are established respectively. The two methods integrate the subjectivity of the evaluation indexes and the objectivity of the traffic participants, which unifies the qualitative classification and the quantitative evaluation of traffic state.4) A traffic state categories prediction model and a traffic state index prediction model are put forward to obtain the two-side of future traffic state.By considering the category prediction as a pattern recognition problem, a category prediction model based on maximum entropy is established, which can integrate the spatial and temporal influencing features of traffic state without considering their relationship. According to the multi-model modelling ideology, a self-adaptive weight combination model is proposed to forecast the traffic state’s index. The model is established on the traffic state category prediction, which can be regarded as a two-stage prediction model. The advantage of the model is it could reduce the uncertainty, adapt to the random changes of traffic flow, and improve the prediction accuracy.5) According to the actual traffic management demands of Beijing city, and based on the above theoretical methods, the Beijing Regional Traffic State and LOS Evaluation System is developed and applied. The successful application of the system verifies our achievements are of validity and practicality.