Short-term Traffic Flow Prediction Based On Hybrid Algorithm Optimized Wavelet Neural Network

With the development of social economy,in most big cities,traffic congestion has gradually become a serious problem,leading to traffic chaos and environmental pollution.Intelligent Transportation System(ITS)is a scientific means to solve this problem.The short-term forecast of traffic flow is always a research hotspot.This thesis firstly analyzes the chaotic characteristics of traffic flow.The phase space technology of reconstructing the traffic flow time series is used to extract the hidden information in the sequence and to recover the attractors of the chaotic time series.Then the thesis uses the C-C method to calculate the embedding dimension and delay time for reconstructing the phase space,and then reconstructs the phase space of the real short-term traffic flow data collected on the PeMs capability measurement system.Furthermore,the small array method in chaos theory is applied to calculate the maximum Lyapunov exponent to analyze the predictability of traffic flow sequences.Secondly,after analyzing the advantages and disadvantages of various short term traffic flow prediction models proposed in recent years,the wavelet neural network(WNN)is selected as the prediction model to realize short-term traffic flow,due to its better prediction accuracy.Compared with the classic BP neural network model,it is shown that the realized short-term traffic flow prediction based on wavelet neural network can achieve the better performance in terms of the prediction accuracy and the convergence rate.Thirdly,in order to overcome the deficiency in getting trapped in local optimum and the undesired oscillation towards the optimal solution caused by improper initial settings of wavelet neural network,a short-term traffic flow prediction model based on PSO-WNN is studied in this thesis.PSO is a kind of group intelligence evolutionary algorithm,which has advantages in convergence and optimization performance.The initial parameters of wavelet neural network can be optimized by the widely divergent dynamic behavior of particles.The simulation results show that,the short-term traffic flow forecasting model based on PSO-WNN is better than the WNN model only.Finally,as the number of iterations of the algorithm increases,the convergence performance of PSO tends to be slower and the optimization ability decreases.To overcome this shortcoming and improve the prediction accuracy,this thesis uses quantum particle swarm optimization algorithm based on quantum mechanics.The global convergence is better than the PSO algorithm,and the cross mutation operation of genetic algorithm is introduced.The issue of single species of new particles in QPSO algorithm is improved.The short-term traffic flow prediction GQPSO-WNN model based on phase space is established and compared with the PSO-WNN-based model and the QPSO-WNN model based on phasespace.It is verified that,there is noticeable improvement in terms of the realized prediction accuracy and the convergence rate of the GQPSO-WNN mode based on phase space,which can therefore be effectively applied in real traffic prediction applications.The contribution of the thesis provides a feasible research idea and a technical solution for the optimized wavelet neural network based short term traffic flow prediction,which is of some value for future research and applications in this area.