The Research Of Convolutional Neural Network-based Large-scale Short-term Traffic Flow Prediction Method

With the rapid development of the urban urbanization process,the number of urban motor vehicles has increased significantly,and the urban road pressure has also increased,which leads to a series of social problems,such as traffic congestion and traffic accidents.The development of intelligent transportation to solve the problem of traffic congestion has become the main strategy,which has promoted the construction of Intelligent Traffic System(ITS).Among them,efficient,accurate and real-time traffic flow prediction is a key element in the establishment of ITS.Especially,short-term traffic flow prediction has special important guidance for urban population on route planning and transportation department on traffic management and guidance.Research on short-term traffic flow prediction theory and methods is of great significance to maximize the performance of urban road networks,and has become one of the research hot-spots in the field of intelligent transportation.In recent years,there have been many studies on short-term traffic flow prediction at home and abroad,and various models for traffic flow prediction have been proposed,but there still exist some problems that need to be solved urgently.This paper conducts research on the above problems,and the specific contents include:(1)A dynamic traffic flow prediction technology based on incremental learning is proposed to solve the problem of high time efficiency and strong uncertainty in large-scale short-term traffic flow prediction.First,clustering algorithm is used to extract traffic accident information from traffic flow data,which is introduced into the model together with the weather and holiday information as uncertain features to improve prediction accuracy.Then the incremental training method is used to replace the traditional batch training method to improve the update speed of model.Finally,the active learning method is used to fine-tune the model to improve its prediction performance in special traffic environments.(2)A short-term traffic flow prediction algorithm based on 3D convolutional neural network(3D CNN)is proposed to fit the spatio-temporal features of traffic flow data and solve the problem of missing data.3D CNN branch is used to process historical traffic flow data with obvious spatio-temporal features and 2D convolutional neural network branch is used to process uncertain features in the model.In addition,the 3D CNN is used to build missing data completion model to improve the quality of the dataset to further improve the accuracy of the prediction model.(3)Based on the above research of key theory and algorithm,a traffic management and control system based on traffic flow prediction,UrbanTFlow,is developed.Through real-time display of the traffic flow situation in a city,the system is verified to be simple and easy to operate.And the technology of traffic flow prediction is practical and effective,which provides a basis for intelligent traffic control and guidance.