The Research Of Short-term Traffic Flow Prediction Base On The Wavelet Neural Network

The guidelines in Guangzhou’s12th “Five Years Plan ", the first majordevelopment strategy is to accelerate the transition, construct the low-carbon,wisdom and Guangzhou. Includes not only the city traffic management ofinformation, intelligent and also the city traffic in order to improve and the citizensconscientiously observe traffic rules for the promotion and realization of the overallquality of the city road traffic jams. The goal of Guangzhou Intelligent TrafficManagement System (ITMS) is intelligent scheduling, scientific management,enforcement automation and the public service diversification. The dynamicanalysis and short-term prediction of traffic flow in Intelligent Traffic ManagementSystem is the very important part. The research content of this paper is fromGuangzhou Intelligent Traffic Management System.City road traffic system is a constantly changing and complicated system, sothe operation is very hard to predict. To achieve better results of experiments, wemust handle the traffic flow short-term prediction. The research content of thispaper is the traffic flow short-term prediction. The research direction is the use ofroad traffic flow dynamic analyses technology on the collection of historical dataand present traffic flow, so that we can predict traffic flow in precise and real-time.The Research goal is intelligent scheduling, scientific management, enforcementautomation and the public service diversification.In this paper we predict the short-term traffic flow by the wavelet neuralnetwork. We analyses the results of experiments, so we sum up the advantage andthe disadvantage of the wavelet neural network’s prediction. The algorithm ofparticle swarm optimized PSO-WNN based on wavelet neural network is presentedto predict the traffic flow, PSO algorithm is applied to optimize the initial weightsof nodes in wavelet neural network, historical data are used to train neural network.The PSO-WNN algorithm speeds up the convergence rate of wavelet neuralnetwork and the accuracy of the traffic flow prediction is improved. Some time theerror of the prediction is unacceptable, so we use the Mean Absolute Percentage Error to monitor the PSO-WNN prediction model. If the MAPE reach the superiorlimit we will re-train the PSO-WNN prediction model. The MAPE monitorsPSO-WNN make sure that the error will all acceptable and have a real-timeprediction. The prediction of the traffic flow can be used in many place in ITMS,we introduce the application of traffic flow prediction in this paper at last.