The Research Of Short-term Traffic Flow Prediction Methods

Urban traffic problems plagues the development of the city for a long time, Intelligent Transportation System (ITS) as a solution to this problem is the most effective means and achieves worldwide recognition. ITS deals with traffic problems mainly from time and space of traffic planning properly.On time control it has traffic control subsystem, on space control it has traffic inducement subsystem, and auxiliary transportation subsystem, etc.. Traffic information prediction techniques is one of the core technology throughout each subsystem.Traffic flow prediction of traffic information is the main direction of this research. Because of the strong nonlinearity and randomicity, establishing the mathematical model of short-term traffic flow is very difficult, this paper attempts to study the intelligent algorithm of many kinds, obtained very good result.Firstly, grey system in chaos theory is chosen, basic conception and relevant theories of modeling of it are briefly discussed; and the grey prediction model GM(1,1) is used as model of traffic flow modeling, the modeling steps and application range are described in detail, the causes of the initial prediction accuracy being bad is analyzed, and on this basis put forward the improvement direction of gray prediction model, results are improved effectively.Secondly, for the grey forecast model robustness is not high, the neural network is chosen as the traffic flow prediction model, which has the expertise in the nonlinear fitting. This paper adopts topology three layer perceptron network, the BP algorithm is joined, to analyze the prediction results, to discusse the number of neurons in the hidden layer and other parameters effect on the prediction, at the same time, more advanced RBF neural network and GRNN neural network are researched, and make a comparative analysis on BP neural network.Thirdly, aiming at some defects on the inherent neural network algorithm, such as the global search ability is not strong, easy to fall into local minimum value, etc., respectively mixed with genetic algorithm and wavelet algorithm of intelligent algorithm are used to make up. Combined with genetic algorithm, the topology three layer perceptron network is chosen, BP algorithm and genetic algorithm to choose the optimal use of parallel computing, the ability of global searching ability of genetic algorithm and the iterative evolution choose the weights and thresholds, and then back to the BP algorithm to do the reverse calculation error, replace the original algorithm in gradient descending part, make the prediction accuracy improved. But the calculating of the genetic algorithm process is complex, large amount of calculation, calculation time is long, demand the higher hardware, So three layer perceptron network topology structure based on wavelet neural networks is chosen. Hidden layer is set to the wavelet basis function, so that the original threshold wavelet parameters and weights will be replaced with the corresponding implicit layer to information processing ability was strengthened, make up for the deficiency of the neural network.Finally, the model has been trained and simulated with the sample of the real short-term traffic flow data in the Yinbin Avenue of the Jiangmen city. The simulation results show that GM(1,1) prediction model has simple algorithm structure, and its accurary can meets the requirements of short-term traffic flow. Genetic neural network and The Wavelet neural network model inherits the strong learning and training ability of BP neural network. It improves the forecast model precision at the same time.Several different prediction models established in this paper each has advantages and disadvantages, but each has its own scope of application.In a complete regional traffic control system, strategic prediction model considering various factors can be applied to the genetic neural network model which has strong calculating ability.Concrete road and intersection can be applied to the wavelet neural network model which is more flexible. In different applications of each algorithm can give full play to their own advantages, and has strong practicability in the intelligent traffic control system.