Research On Prediction Algorithm Of Spatiotemporal Road Condition Based On Graph Neural Network

With the continuous development of science and economy,more and more people begin to have higher requirements on the way of travel and commuting time.The emergence of online car hailing mode enriches people’s way of travel.It not only provides convenience for people,but also brings negative impact on urban traffic congestion.On the one hand,the increasing number of urban population puts constant pressure on the existing urban transportation infrastructure.On the other hand,the possession of passenger cars is getting higher and higher.The model of online car hailing enables some people to flexibly find jobs as private car drivers,which leads to more and more passenger cars on the road.Therefore,it is an important part of intelligent transportation to accurately predict road congestion,provide accurate prediction results for drivers and people and recommend unimpeded roads.For ride-hailing platforms,providing fast and congestion-free roads for drivers can,on the one hand,improve the efficiency of drivers receiving orders and reduce operating costs,and on the other hand,provide users with good ride-hailing experience and improve users’ adherence to the platform.In recent years,due to the continuous improvement of computing power such as GPU,deep learning technology has started to rise again.At present,deep learning technology is often data-driven and needs to learn a large amount of data.Meanwhile,due to the natural richness of traffic data,deep learning technology can be well applied to traffic data.In the field of traffic big data,deep learning technology has shown excellent performance.In real life,there are a lot of structured data,such as people’s social networks and urban traffic road networks.Graph neural network is a neural network applied to graph structure data.Traffic road network has a natural graph topology.Therefore,compared with other neural networks,such as convolutional neural network,graph neural network can better extract spatial features of graph structure data.Based on the open source data set provided by the ride-hailing platform Didi,this paper proposes a prediction algorithm for spatiotemporal road conditions based on graph neural network.In the traffic road network,the influence of upstream and downstream adjacent roads on the central road is different.Because the traffic flow is directional,the influence of upstream adjacent roads on the central road is greater than that of downstream adjacent roads on the central road.Therefore,this paper first improves the graph neural network.The graph neural network is more focused on the spatial information of upstream adjacent roads when extracting map spatial features.The output results of the graph neural network are fed into the recurrent neural network and trained with the time series data at the same time,so that the model can learn the temporal and spatial correlation of traffic data.Finally,multi-channel convolutional neural network is used to carry out convolution operation on historical periodic data to fully mine the cross-features of road condition information and cross-features of road condition information across time scale.Since the traffic data is non-linear and unstable with time,and its spatial correlation also changes with time and is dynamic,the influence of adjacent roads at different times on the central road is also different.Based on this,this paper proposes a neural network structure based on attention mechanism,which can learn the spatial feature vectors of time slices to be predicted according to the historical high-order spatial feature vectors,and provide more comprehensive and precise spatial features for traffic congestion prediction.This paper conducts experiments on the open source data set provided by Didi,and selects other works based on statistical learning,machine learning and deep learning for experimental comparative analysis.The neural network structure proposed in this paper has achieved excellent performance in traffic congestion prediction of ride-hailing.