The Study On The Real-time Identification Of Chaos In Traffic Flow

Traffic flow system is a human-joined, changeable, open and complex huge system. It is high nonlinear and uncertain. Under certain condition, chaos appears in it. The summary of studies on chaos in traffic flow suggests that study on chaos is of great theoretical and practical value. In addition, the summary of methods to identify chaos shows the limitation on current methods and the direction and angle of study on real-time identification of chaos in traffic flow. In this thesis, these are two approaches to real-time identification of chaos in traffic flow. The first one is the combination of improved small-data method and improved surrogate-data technique. Small-data method is easy to compute, antinoise and reliable for small data. Surrogate-data technique can avoid false identification. However, the novel approach has not only the advantage of small-data method, but also the rigor of surrogate-data technique. In this approach, small-data method and surrogate-data technique are all improved and its computing steps are first introduced in detail. The case studies of this approach are given for time series of traffic flows generated by Bierley Car-following model and microcosmic simulation software and real vehicles. The second one is based on the thought of finding the relationship of chaos and the initial condition. Based on the conception and methods of data mining techniques, an intelligent system framework for real-time identification of chaos in traffic flow is proposed. Furthermore, the function and implementation of every module in the system is introduced, the principle and steps of the arithmetic is proposed and the selection of methods is discussed and explained. The case studies of this system are given for time series of traffic flows generated by Logistic system and microcosmic simulation software and real vehicles. Moreover, when other experimental condition and steps are same, the influence of different factors on identification results is discussed. These factors include feature of time series, such as power spectrum, wavelet packet coefficients and wavelet packet energy, the length of time series, layers of wavelet packet decomposition, wavelet function, algorithms of knowledge discovery, such as BP neural network and Fast Classification for Support Vector Machines. The results show the selection for factors. All above experimental results indicate that these two approaches to real-time identification of chaos in traffic flow are robust and can be fit for requirements of real-time performance and veracity.