Research On Short-term Traffic Flow Forecast Based On Hybrid Neural Network And Multidimensional Factors

Traffic congestion,as one of the most important problems to be solved in the world,has been paid more and more attention by governments.Intelligent transportation system is another innovative way to solve the problem of transportation in today’s society.It applies advanced science and technology to the road traffic system,aiming at solving the problem of traffic congestion from many aspects such as monitoring,warning and inducing.Traffic flow prediction is the most important part of the intelligent transportation system.Efficient,real-time and accurate traffic flow prediction can effectively alleviate the urban road conditions.For one thing,it can provide reliable guidance information for People’s Daily travel and help to reduce the possibility of peak travel.For another,it can also provide more reliable decision-making basis for the government and other road traffic management departments to realize the optimal design of urban roads.The existing traffic flow prediction method and theory has achieved good results.Given the traffic flow is affected by many factors of complex time series and hybrid network structure strong ability of feature extraction,and the research is relatively small for these aspects of.Therefore,started with the basic theory of traffic flow,the research studies the research of traffic flow prediction model.The research ’s main work is as follows:1.It discusses the basic concept of traffic flow,analyzes the basic characteristics of traffic flow and explores the influence of each characteristic on traffic flow.And then,it reviews the classical traffic flow prediction theory and discusses the existing problems and processing steps of traffic flow data preprocessing technology.2.Considering that the traffic flow is affected by weather factors and the traffic flow data presents the characteristics of sequence data,a traffic flow prediction method based on weather weighted factor and queue hybrid neural network is proposed.Correlation coefficient method is used to build the weather weighted factor model,and queue unit is introduced to build the hybrid neural network.Finally,15-minute data sets are used to train and predict the model.Compared with similar T-LSTM,T-GRU and T-Bi-LSTM models,the model proposed in this research is superior to other models in accuracy.3.In view of the strong spatio-temporal characteristics of traffic flow and the advantages of Bi-LSTM network and Conv-GRU network in temporal and spatial feature extraction respectively,a traffic flow prediction theory based on spatio-temporal analysis and mixed deep network is proposed.Firstly,the periodic similarity of traffic flow is analyzed quantitatively,and the adjacency matrix of road network stations is established by using graph theory.Then,the Bi-LSTM network is used to extract the temporal periodic characteristics of traffic flow,and the Conv-GRU network is used to extract the spatial characteristics of traffic flow.Finally,the result of feature extraction is fused.The experimental results show that this model has obvious advantages in spatiotemporal feature extraction.It improves the accuracy of prediction effectively compared with the existing relevant models.