Research And Implementation Of Short-time Traffic Flow Prediction Based On Chaos And Particle Swarm Optimized ANN

Traffic flow prediction is an important research area of intelligent transportation system. In the Intelligent Transportation System, fast and accurate prediction of traffic flow is an important premise and foundation of applications in real-time traffic signal control, traffic distribution, route guidance, automatic navigation, incident detection, and etc. However, transportation system is a complex system composed of human, vehicles, roads and many other factors, which with characteristics of highly complexity, nonlinearity, uncertainty. So accurate, real-time, reliably traffic flow prediction has become an important research.According to the principle of predictability analysis and the theory of chaotic time series analysis based on traffic flow dynamics properties, this paper analyzed the real traffic flow data collected from PeMS in California USA. The phase space reconstruction technique is used in the traffic flow prediction, expected to explore the laws embedded in the traffic flow to enhance traffic flow prediction accuracy. Chaos of the traffic flow is analyzed with C-C method and small data sets. The C-C method is used to access the phase-space reconstruction time delayτand embedding dimension m. After the phase-space reconstruction on the traffic flow time series data, we use the method with a small dataset to gain the non-linear chaotic characteristics of the Lyapunov exponent, which is greater than zero verified the existence of traffic chaos.Traffic flow data has characteristics with highly complexity and nonlinearity, and artificial neural network has a very strong ability with non-linear processing, self-organizing, self-adaptively and self-learning, which is an effective way in traffic flow prediction. Particle Swarm Optimization Algorithm is a new evolutionary algorithm, with high convergence speed, high robustness, strong global search ability, and no requirement of the problem's feature information. In this paper, Particle Swarm Optimization Algorithm is used in BP neural network as training algorithms to build PSO neural network. This paper improved the PSO algorithm in aspects of Inertia weight, speed limits, convergence properties, search capabilities, and etc. The improved algorithm promoted PSO neural network's convergence speed, training accuracy and generalization. Real-time and accuracy of Short-term traffic flow prediction with PSO neural network was also assured by it.Finally, a single-point & single-step short-term traffic flow prediction model with PSO Neural Network in chaotic phase space has been set up. Research using authentic traffic flow data collected from PeMS has been divided into two parts, workdays'regularity and holidays'irregularity. Prediction results of PSO neural network model have been compared and analyzed, from before to after comprehensive improvement of Particle Swarm Optimization Algorithm. The results fully validate the effectivity of improved PSO algorithm and single-point & single-step traffic flow prediction model with PSO Neural Network in chaotic phase space.