Research On Short-Term Traffic Flow Forecasting Method Based On Chaos And Wavelet Neural Network

With the continuously development of intelligent transportation system, research of the analysis and processing in road traffic flow have also launched, through the prediction of future traffic flow information accurately in real time, the State and the Government can find appropriate control strategy to improve the traffic congestion, to make the road network unobstructed and operation efficiently. So, it has great social significance to research the short-term traffic flow prediction methods of city traffic system.Based on the research of the chaos characteristics in traffic flow, the phase space reconstruction can revealed the implicit information of one-dimensional the traffic flow time sequence, then describe the chaotic attractor. This article uses the C-C method to calculate the required embedding dimension m and time delay τ of phase space reconstruction of the real traffic flow data in USA California PeMS system, and by using small data sets method to calculate the largest Lyapunov exponent to identify the existence of chaos, and then using the chaos theory to analysis and prediction research the traffic flow sequence.Based on the analysis of the advantages and disadvantages of various traffic flow prediction model, choosing the wavelet neural network (WNN) as the short-term traffic flow prediction model, and chooses the actual traffic flow to do the simulation experiment by analysis and comparing the BP neural network model and the WNN model, verified the WNN model for prediction of short-term traffic flow both in accuracy and convergence are better than BP neural network prediction model.Aiming at the instability and the slightly higher prediction error in wavelet neural network, the genetic algorithm (GA) is introduced as the optimization method of weights and wavelet parameters in WNN network, make up for the defects of sensitive to initial value in WNN network, improving the algorithm through three aspects which is improving the coding algorithm, genetic operator’s self-adaptive transform, and combining with ant colony algorithm, makes full use of the global search ability of the genetic algorithm and local search ability of the ant colony algorithm together to predict the short-term traffic flow. Through comparing of comparative prediction performance index in simulation experiment proved improved IGA-WNN model better than GA-WNN model and WNN model in precision and stability, improving the real-time online short-term traffic flow prediction’s applicability.