Study On Key Theory And Methods For Nonparametric Recognition Algorithms Of Traffic Flow

Traffic flow condition recognition is one of the important issues for Intelligent Transportation Systems, especially for the Advanced Traffic Management Systems and Advanced Traveler Information Systems. It mainly includes the recognition of quantitative change (the distribution of traffic flow is invariant, but the parameters change), real-time recognition (namely incident detection or traffic flow breakdown detection) and traffic flow recognition in advance (namely traffic flow forecasting). Its main function is to analyze road condition through the study of traffic flow forecasting and incident detection, evaluate the dynamical traffic information deeply, and recognize the future and real-time traffic under different condition timely and accurately. The key part of the system is the fulfillment of core algorithms, which is the main research content of this paper.Combining the National Intelligent Transportation System Architecture, based on the basic thought of dynamical traffic management, a framework of traffic flow condition recognition systems is constructed. Aimed at the basic content of traffic flow condition recognition systems, the key theory and methods for nonparametric recognition of traffic flow are studied in this paper, such as the short traffic flow forecasting model based on projection pursuit auto-regression, nonparametric probability change-point model in traffic quantitative change, the traffic incident detection based on support vector machines and the traffic fuzzy information fusion based on genetic algorithm. These models are validated with the field data and simulation data from the traffic simulation system software of VISSIM. The results indicate that these models and algorithms in the paper have better performance, and provide new research methods for traffic flow condition recognition and traffic flow guidance.