Short-term Traffic Flow Prediction Based On Neural Network And Particle Filter

With the rapid advancement of urbanization in China and the increase in the number of motor vehicles,urban road networks have taken an important role in residents' travel and social resources deployment.Short-term traffic flow forecasting has always been an important issue in urban road traffic,and is the basis for managers to accurately and effectively manage road traffic conditions.Accurate short-term traffic flow prediction can effectively predict the status of traffic flow at the next moment,provide travel guidance for the traveler,avoid congested road sections,quickly reach the travel destination,ensure the safety of the road network,and thus make the city road network in the efficient operating conditions.Firstly,this paper summarizes the research results of many scholars at home and abroad,summarizes the features of short-term traffic flow prediction,and classifies commonly used forecasting methods.It selects typical algorithms for each type of traffic flow forecasting method and briefly describes its principles and discusses Its advantages and disadvantages.The BP neural network and particle filter algorithm are selected as the basis of this study.An improved algorithm is proposed to use the BP neural network prediction process as the state transfer equation of the particle filter algorithm,so as to realize the dynamic optimization of the structural parameters of the BP neural network algorithm.The algorithm can avoid the problem of over-fitting to a certain extent,and is more sensitive to changes in the state of traffic flow,and thus has better real-time prediction effects when the state of traffic flow changes.Finally,this paper selects the traffic flow average velocity time series of one road in Guangzhou as the experimental data set of this paper,and verifies the prediction effect of the improved algorithm.Wavelet threshold denoising and phase space reconstruction are used to process one-dimensional velocity time series data and extend it into high-dimensional space.The effect of BP neural network algorithm and improved algorithm on short-term traffic flow prediction is compared.The related predictive evaluation indicators show that the improved algorithm has better accuracy and robustness.