Research On Vehicle Travel Path Recommendation Method Based On Graph Neural Network

With the increase of urban permanent population and vehicle ownership,the traffic congestion problem in some cities is becoming increasingly serious.Urban traffic congestion is frequent and the traffic status is complex,especially the growth rate of vehicles far exceeds the growth rate of transportation facilities and road construction.Traffic congestion has brought huge challenges to the sustainable development of the city.In order to ensure smooth traffic,reduce traffic congestion,improve road safety,and reduce the negative impact of air pollution on the environment,traffic management departments are turning to scientifically developed new technologies.Path recommendation is of great significance in improving traveler travel efficiency,reducing travel costs,balancing road network operation status,and reducing time and space congestion.This article analyzes the characteristics of checkpoint data,excavates the spatiotemporal characteristics of vehicle travel time in urban road networks,analyzes regional traffic status by dividing the urban road network,and proposes a path recommendation algorithm based on vehicle travel time.Under the premise of considering road network balance,by improving the accuracy of dynamic travel time prediction and dynamically considering path evaluation,the path recommendation method for urban vehicle travel is optimized.The main content is as follows:A road travel time prediction model based on dynamic graph convolutional network.This method utilizes the correlation between the travel time data of the current road segment and not only the historical spatiotemporal data of the current location,but also the travel time of adjacent road segments.Combining with the topological structure of the road network and utilizing spectral filtering,the model extracts spatial features of travel time,thereby improving the accuracy of road segment travel time prediction;The dynamic Laplacian matrix is introduced to extract the spatiotemporal characteristics of the road network,and a dynamic graph is constructed for the graph convolution neural network.The graph time domain convolution layer is designed to extract high-dimensional local time domain information from the original traffic data.In the time attention mechanism,time attention comes from adaptively capturing the large-scale temporal correlation of traffic data.Batch＿Norm is the activation function of travel time prediction based on graph convolution network,outputs the spatial and temporal characteristics based on historical travel time,and constructs the loss function of the model through normalization.Based on the weighted GN(Girvan Newman)algorithm with travel time as a weight,a sub district partitioning model for urban road networks is proposed.Based on the topological structure of the road network,checkpoint points are used as nodes and road network sections are used as edges.By dividing the edge intermediate number by the corresponding edge travel time as a normalized weight,a road network weighted graph is constructed using the time series of road travel time.The modular function is used as the partitioning result to measure the poor cohesion of unauthorized community division.Furthermore,considering the strong correlation between adjacent road sections in actual travel time,the problem of poor cohesion in unauthorized community division is addressed,more accurate dynamic real-time state area division of the road network;Combining the average travel speed indicator in the "Evaluation Method for Traffic Congestion Degree" to identify traffic congestion degree.Based on the congestion status of different regions and the predicted travel time,a vehicle travel path optimization recommendation model considering road network balance was established.Combining the predicted travel time of the dynamic graph convolutional network with the topology of the road network,a road network travel time matrix is constructed.The shortest path algorithm is used to obtain the shortest travel time matrix and corresponding routing matrix between all checkpoint points in the road network;In order to further characterize the probability of congestion risk,this article introduces a congestion risk coefficient and combines it with predicted travel time to form a new road network travel time matrix.It constructs a path selection recommendation algorithm based on road network equilibrium.Combining the congestion risk coefficient,it considers whether to avoid heavily congested areas and compares the travel time and road network congestion situation of path planning under different path recommendation conditions.Based on this,path selection recommendations for congestion avoidance can effectively reduce the congestion equilibrium index of the road network,and consider the congestion avoidance mechanism from a global perspective to alleviate the overall congestion level of the road network.