Traffic Flow Prediction Based On Graph Convolution Network And Information Geometry

Under the background of the significant progress of automobile industry and the increasing improvement of residents’ economic level,the number of cars in China continues to increase,which brings traffic congestion,environmental pollution and other problems are becoming increasingly serious.The Outline for Building China’s Strength in Transportation issued by the CPC Central Committee in conjunction with the State Council clearly points out that intelligent,digital,lightweight,environmental protection and sustainable development are the main directions of transportation construction in the future.To develop a green and efficient intelligent transportation system is a national strategy and the aspiration of the people.Traffic flow prediction is an important sub-subject in the intelligent transportation system.Traffic flow is an important data to be analyzed by the intelligent transportation system.The regional traffic flow can well reflect the current traffic conditions in the region.Using historical data of traffic flow to analyze the potential laws of traffic flow and forecast the future traffic flow has significant practical value.However,effective analysis of highly complex and nonlinear traffic systems is a challenging topic.In view of the highly spatiotemporal dynamics of traffic data,a new traffic flow prediction depth model is proposed based on graph convolution and information geometry.The graph convolution model uses information geometry method to describe the distribution difference of traffic node data and forms a dynamic matrix combined with attention mechanism.And it is applied to the graph convolutional network which uses fuzzy relation to model the transfer relation of traffic data to extract the dynamic characteristics of traffic flow data effectively.The experiment shows that it has more accurate prediction effect.The main work of this paper is as follows:1.Data processing and traffic flow data set formation.There are different ways to collect traffic flow data in industry.Aiming at two typical ways,this paper processes the collected real traffic data in different forms from many aspects,such as data denoising and data complement,and finally forms a usable traffic data set with the same form.2.Design and Implementation of Traffic Flow Prediction Model.The main task of the traffic flow prediction model is to extract the temporal and spatial characteristics of the traffic system effectively.In this paper,graph convolution is used to extract the spatial topology of traffic network effectively,and multi-layer cause-and-effect void relation is used to extract the temporal dimension features.At the same time,considering the dynamic characteristics of time and space in the traffic network,this paper uses the attention mechanism to model the temporal and spatial dynamics,and strengthens the attention matrix by considering the difference of data distribution among traffic nodes through the information geometry method.In the graph convolutional network,the fuzzy relation theory is used to describe the transfer relation between traffic nodes,and the attention matrix is used to capture the transfer relation between traffic nodes dynamically,so as to achieve more accurate prediction effect.Finally,the validity of the prediction model is proved in a variety of real world traffic data sets.3.Based on the prediction model,a real-time road condition evaluation method is designed and constructed.In order to transform the prediction results of the deep traffic flow prediction model into effective cognition,this paper constructs two indicators for measuring traffic flow based on historical data,and then uses the fuzzy comprehensive evaluation method to generate the road condition evaluation of traffic nodes.