Research On Short-term Traffic Flow Predication Based On Integrating Neural Network

With the increasingly serious problem of urban road traffic congestion,intelligent transportation systems have emerged to effectively improve road capacity at the historic moment.However,the effective implementation of intelligent transportation system requires accurate and timely short-term traffic flow data as support.Accurate and timely traffic flow prediction not only helps to formulate effective traffic management and control strategies,reduces waste of right of time and space,and improves transportation efficiency;but also can provide travelers with accurate and reliable route guidance information,so that traffic flow has a reasonable distribution on the road network.Based on the existing research results on short-term traffic flow prediction at home and abroad,and combining the characteristics of short-term traffic flow,this paper proposes an integrating neural network model based on secondary decomposition technology.What's more,the paper focuses on two aspects of preprocessing technology of short-term traffic flow data and optimization of prediction models.The main contents done are as follows:(1)Preprocessing technology of short-term traffic flow time series based on quadratic decomposition method.Firstly,the wavelet decomposition method is used to reduce the noise of the traffic flow to reduce the interference of the noise on the prediction of the traffic flow.A wavelet decomposition method with a relatively high signal-to-noise ratio is finally obtained based on a large number of simulation experiments.Then,the ensemble empirical modal decomposition is performed on the signal after wavelet decomposition.The traffic flow is decomposed into several waves with similar frequencies and residual wave to improve the accuracy of the prediction model.(2)PSR-PSO-Elman neural network prediction model.Due to the chaotic nature of the traffic flow,the phase space reconstruction technique PSR based on the C-C method was used to determine the number of input layer nodes in the Elman neural network structure.The model uses multiple hidden layers instead of single hidden layers to improve the efficiency of network feature recognition by hierarchically averaging the best hidden layer nodes.Finally,the particle swarm optimization was used to optimize the connection weights and thresholds of the Elman neural network,and the PSR-PSO-Elman neural network prediction model was finished.(3).Decomposition-Ensemble Elman neural network prediction model.Each decomposition component with the same or similar characteristics is predicted separately by the PSR-PSO-Elman neural network model,and then the prediction results of each component are added and integrated to obtain the final integrated neural network prediction result.The model proposed in the paper was verified by some short-term traffic flow data of three detection stations in the Freeway Performance Measurement System PeMS in California.The experimental results show that the prediction accuracy of the integrating neural network model is higher than that of the single model,and the average relative errors decrease from 9.80% to 6.97%,from 8.83% to 6.42%,and from 7.82% to 5.71%.