H.Q. Zhang, C.H. Wang

Pages: 109-116

Abstract
The contradiction between nonlinearity and complexity of traffic flow change and real time and accuracy of traffic flow forecasting drives the study on relevant optimization algorithms. To provide intelligent traffic system with the better data and decision support, it proposes the improved wavelet neural network forecasting model of particle swarm optimization algorithm and cloud particle swarm optimization algorithm. Based on the study on short-term traffic forecasting and the theoretical exposition to wavelet neural network, this paper analyzes and summarizes such disadvantages of the wavelet neural network as slow convergence rate in the aspect of traffic flow forecasting; it improves the wavelet neural network forecasting model by utilizing the advantages of high convergence rate and strong global searching ability of the particle swarm optimization algorithm; and it gives contrastive analysis to convergence rates and forecasting error of various models by conducting simulation experiment. Finally, it is indicated by the results of simulation experiment that: the improved model of cloud particle swarm optimization algorithm possesses the highest traffic flow forecasting ability, followed by particle swarm optimization algorithm and wavelet neural network algorithm successively, which illustrates the effectiveness of modified model. The study in this paper provides the better technical support and theoretical guidance to the optimization of urban intelligent traffic management system.
Keywords: wavelet neural network; particle swarm optimization algorithm; cloud particle swarm optimization algorithm; traffic flow forecasting; simulation experiment