Study On Traffic Flow Prediction And Optimal Control Of Urban Road Network Based On Spatiotemporal Feature Fusion

With the rapid development of China’s economy and increasing demand for transportation,the traffic load and pressure are increasing sharply.It causes traffic problems such as traffic congestion,traffic accidents,air pollution,fuel consumption,noise pollution and affects travel efficiency and economic development seriously.By studying road network short-term traffic flow prediction techniques can obtain the roads future traffic status in advance,conduct dynamic division of urban road network subregions,and optimize road network signal control strategies.It can reduce delay time,queuing length and parking times,increase road capacity,improve road network operation efficiency,and ensure traffic safety and order,having good economic and social benefits.Firstly,this paper analyzes and processes urban road traffic flow data,mines the spatiotemporal characteristics of traffic flow data fully,builds short-term traffic flow prediction models using 3D convolution,gated recurrent networks,residual networks,integrates traffic flow data,weather,holidays,weekends and other factors to predict future short-term road network traffic flow.Compared with the baseline models,the model in the paper reduces RMSE by 4.86%and 9.10%on two public datasets,respectively.It improving the accuracy of traffic flow prediction and laying the foundation for dynamic division of road network subregions.Furthermore,an intersection correlation model is established based on traffic correlation coefficient,density correlation coefficient and signal cycle correlation coefficient.It improves the traditional Newman community division algorithm for adapting to the division of undirected and weighted road network subareas.A dynamic subarea division method is proposed with distance principle.Through simulation analysis of the core area road network in Lixia District,Jinan,the average dispersion of the road network for the division method in this paper is 0.4476,which is the lowest compared to the traditional Newman algorithm and the improved Newman algorithm.It verifies the effectiveness and rationality of dynamic partitioning of road network subareas.Finally,a road network signal timing optimization model is constructed with the optimization objectives of delay time,parking times and queuing length for each divided traffic subarea.The fuzzy matrix is used to allocate the target weights dynamically.An improved hybrid particle swarm optimization genetic algorithm is used to solve the parameters of the timing model.In the VISSIM-Visual C++joint simulation control platform,the core area of Lixia District,Jinan,including 12 key intersections,is simulated and analyzed to achieve optimal control of traffic flow in the road network.The delay time,parking times,and queue length in this simulation area reduces by 11.05%,18.17%,and 15.87%,respectively,improving the traffic capacity of the road network greatly.In summary,based on the short-term traffic flow prediction model of the urban road network,the paper implements the dynamic subarea division of the road network based on the improved Newman algorithm.It constructs a multi-objective optimization model for signal timing for each subarea,and uses the improved hybrid particle swarm genetic algorithm to solve model parameters.To achieve traffic flow prediction and control of the urban road network,simulation and analysis are conducted on the core road network area of Lixia District in Jinan in the VISSIM-Visual C++joint simulation control platform.