Research On Prediction Of Urban Travel Traffic Flow And Lane Occupancy Rate Based On Deep Learning

With the acceleration of urbanization,traffic congestion and other problems caused by the development of urban travel have also become obstacles to the rapid development of society.The research and prediction of the traffic flow and lane occupancy rate of urban travel can not only reflect the resource utilization of urban construction and provide reference for urban construction planning,but also provide guidance for people's travel and intelligent travel scheme.The prosperity and development of deep learning provides strong technical support for traffic prediction research.In the previous studies of traffic prediction,the focus is the impact of time and space on traffic prediction,but many factors are ignored.In this thesis,two traffic prediction models are proposed.Firstly,considering the influence of spatiotemporal correlation and lane occupancy rate on traffic flow,a traffic flow prediction model is proposed.Secondly,based on the influence of spatio-temporal correlation and implicit factors of traffic flow and vehicle speed on lane occupancy rate,a multi-component fusion lane occupancy rate prediction model is proposed.The main work and innovation of this thesis are as follows:1.Traffic flow prediction model based on spatio-temporal dilated graph convolution is proposed.In this model,we not only extract the temporal and spatial characteristics of historical time traffic information through the spatiotemporal block composed of the graph convolution and dilated convolution,but also creatively add the lane occupancy information of each node in the model,fuse the extracted characteristics,and output the predicted traffic flow prediction.2.Based on the relationship between lane occupancy rate and traffic flow,a multi-component fusion prediction model for lane occupancy rate is propose.The prediction object of this model is changed to lane occupancy rate.We not only consider the influence of time and space characteristics of traffic volume on lane occupancy rate,but also add traffic flow and speed as implicit factors.It makes the model fuse the components of each feature to predict the lane occupancy.In order to verify the correctness and validity of the model proposed in this thesis,the experiment is based on a data set published by the California Department of transportation,and it is compared with model methods such as HA,ARIMA,LSTM and STGCN.The error rates of MAE,RMSE and MAPE are used as the measurement indexes.The conclusion is that our model reduces the error rate and improves the prediction accuracy by 17%.Although the purpose of this thesis are the prediction of traffic flow and lane occupancy,it also can be extended to the other direction of multi-component prediction based on time and space.