Research On Optimal Control Of Regional Traffic Signals Based On Traffic Flow Prediction

With the continuous development of urbanization and economic level,the number of motor vehicles is increasing day by day,and the traffic congestion problem is seriously affecting the lives of residents.In recent years,the construction of intelligent transportation system has made significant contributions to improving urban traffic problems.Studying traffic flow prediction and regional traffic signal control has important theoretical and practical significance for accelerating the construction of intelligent transportation systems.Aiming at the problem of traffic congestion,based on the existing methods of traffic flow prediction and regional traffic signal control,this dissertation puts forward a traffic flow prediction method based on the spatial-temporal geometric graph convolution network,determines the traffic congestion time and place of regional road network based on the predicted traffic flow data,and optimizes the traffic signal timing in different traffic sub-regions by combining the improved regional road network partition method based on spatial-temporal correlation and artificial bee colony algorithm.The main work and contributions are as follows:1.A traffic flow forecasting method based on spatial-temporal geometric graph convolution network STGGCN is proposed.At present,when using graph neural network to extract spatial features,it is impossible to establish long-term dependence between nodes due to the loss of structural information of neighboring nodes,and the bottleneck of temporal feature utilization due to high complexity in long-term complex forecasting tasks.Geometric graph convolution network and Autoformer are used to capture the spatial-temporal features of traffic flow.Four public traffic datasets are used to verify that STGGCN model has advantages over the other three models in short-term and long-term forecasting tasks.2.An improved traffic network partition method based on spatial-temporal correlation is proposed.Combining deep unsupervised learning with community detection algorithm,this dissertation adopts the fusion model of geometric graph convolution network and temporal convolution network to extract the spatial-temporal correlation of traffic network data.Based on feature embedding representation,the Fast unfolding community detection algorithm is used to divide heterogeneous traffic network into homogeneous traffic sub-regions.Simulation with SUMO software and comparison with other methods verify the superiority of this method.3.The artificial bee colony algorithm based on discrete coding and local search is used to optimize the traffic signal timing scheme in different sub-regions.According to the traffic conditions of different traffic sub-regions,different optimization objectives are used,and multi-objective optimization strategies are adopted for congestion subregions.The effectiveness of this method is verified by simulation with SUMO software and comparison with other meta-heuristic algorithms.Based on the above research results,this dissertation preliminary designs and implements an urban traffic intelligent control subsystem.The subsystem realizes the functions of traffic flow prediction and traffic signal timing scheme.