Research On Highway Traffic Flow Forecast And Traffic State Discrimination Based On Analysis Of Temporal And Spatial Characteristics

At present,traffic managers at home and abroad have insufficient grasp of accurate traffic information,and traffic forecast values are not accurate enough.In order to provide people with better traffic guidance services,new methods must be used to predict traffic flow data with small errors.On the other hand,it is also necessary to accurately determine the traffic state in advance so that people can choose a route with a relatively short travel time.Aiming at the traffic characteristics of California highways in the United States,the paper mainly studies the three-parameter prediction of traffic flow and traffic state discrimination.The specific research contents of the paper are as follows:Firstly,the scientific research background and significance of this paper are given,and the current research status of traffic flow forecasting and traffic state discrimination at home and abroad is described respectively,on this basis,the three elements of California highway traffic flow are discussed in this paper,then the technical route of this paper is put forward.Secondly,the current traffic flow collection and preprocessing technology and the data collection and preprocessing methods of highway are briefly introduced,and the temporal and spatial characteristics of traffic flow are studied.in the time dimension,this paper analyzes the traffic flow characteristics of the same section on different working days in the same week as well as the same section traffic flow characteristics on the same working day in different weeks.In the spatial dimension,the traffic flow characteristics of different sections on the same working day and the traffic flow characteristics of different lanes in the same section on the same working day are analyzed.On the basis of introducing the types and selection principles of traffic state discrimination index,this paper leads to the traffic state discrimination index: traffic flow three parameters(flow,speed and occupancy).Thirdly,a convolution bidirectional long short-term memory network based on hierarchical spatiotemporal attention mechanism(HSTACBN)is proposed,which considers both temporal and spatial factors.The convolution neural network(CNN)which has the ability of spatial local feature extraction is combined with the bidirectional long short-term memory(Bi LSTM)which can simultaneously consider the long-term information of forward and backward directions.A hierarchical spatiotemporal attention mechanism(HSTA)is added at the top letting the network architecture pay more attention to the time and space factors which have more weight to the final prediction,and use it to predict the traffic flow which can better reflect the fluctuation of time and space.In this paper,five benchmark methods are used as comparative models,three error evaluation indexes are used to evaluate the prediction effect of the model,and the training and testing are carried out on the datasets of multiple continuous sections of a certain highway in California,USA.Then,an improved particle swarm optimization k-means clustering algorithm,IPSO-k-means,is proposed to solve the disadvantage of arbitrarily selecting cluster center points in the early stage of k-means clustering iteration,and PSO is used to optimize the cluster centers in the early stage of k-means iteration.Then,the improved particle swarm algorithm(IPSO)is applied to optimize the particle swarm algorithm(PSO)to enhance the overall global search and optimization ability of the clustering algorithm,and the optimal number of clusters k is comprehensively determined to be 5according to the five types of clustering effect evaluation indicators.Finally,an example is given to verify the indirect and direct discrimination of the traffic flow state of the California highway in the United States.Indirect discrimination of traffic state first uses HSTACBN to predict the three parameters of traffic flow,and then inputs the predicted three parameters of traffic flow into IPSO-k-means clustering algorithm to distinguish the future traffic state of different sections of highway.Direct discrimination of traffic state first uses IPSO-k-means clustering algorithm to identify different traffic states,then inputs different traffic states into the prediction model to predict the future traffic states of different sections of the highway,and compares the errors of indirect discrimination and direct discrimination of traffic states.the results show that the recognition accuracy of the two algorithms is higher and has a better application prospect.There are 61 figures,22 tables and 98 references in this paper.