Research On Short-Time Traffic Flow Prediction Based On Improved Wavelet Neural Networks

Traffic flow prediction is the pre-condition for the inducement and control of theIntelligent Traffic Systems(ITS).With strong nonlinearity and degeneration, the trafficflow requires real-time accuracy of forecast models.So real-time,accurate traffic flowprediction has become a hot research.Traffic flow is in fact chaos characteristics instead of seemingly disorganized. Sopredictability of traffic flow should be analyzed firstly.It will be able to ensure thepredictbility of traffic flow if prediction time is less than prediction scale.Optimaltime lag is calculated by the autocorrelation function and optimal embeddeddimension is computed by GP algorithm in the paper. Then the predictability ofdescription analysis of the traffic flow to the recurrence graph image is established.Prediction results will be directly affected by the quality of traffic flow samples. So itis necessary to have preprocessing with traffic flow samples. An improved thresholddenoising method is presented in this paper in order to have preprocessing with thedata of the traffic flow.A quantity of preliminary study shows that it is hard to reach the presumedresults through a simple single algorithm alone. So this paper has been put forward amethod of traffic flow prediction based on improved Ant-Wavelet NeuralNetworks(IACO-IWNN for short). In the early stage, this method the paper posed hasbeen used the improved the “aco algorithm” to train parameters of the “WaveletNeural Networks” in order to get a better solution of global scope. In the late stage,the “BFGS algorithm” is utilized to improve wavelet neural networks, overcoming thedisadvantage of slow convergences of single “Aco Algorithm”.A method of improved aco-algorithm to optimize wavelet neural networks(IACO-IWNN) is proposed. On the one side adaptive pheromone to the evaporationintensity is utilized in order to improve the ability of getting a better solution in globalscope; on the other side, improved pheromone to update formula in order toemphasize the impact of the optimal path on the next generation of ants.This paperhas been taken instance verification for improved ant-wavelet neural networks(IACO-IWNN for short).It is proved that the effectiveness of the IACO-IWNN modelproposed in this paper by predictive simulation of IACO-IWNN model to the trafficflow data of the Yuebei Guangzhou Expressway in15minuets, and by comparing prediction error evaluation index of the corresponding models.