Deep Learning-based Road Traffic Flow Prediction Methods

With the rapid and continuous development of economy and urbanization,the number of motor vehicles has increased steadily year by year.While motor vehicles bring convenience to people’s lives,their rapid growth has also aggravated the problem of road traffic congestion in city.Traffic flow prediction,as an important part of the intelligent transportation system,can not only assist people to formulate more reasonable travel routes,but also help traffic management departments to make more effective judgments on traffic control and guidance.Therefore,it is necessary to study the traffic flow prediction and improve the accuracy of it.In recent years,with the large-scale deployment of traffic detector equipment,the amount of traffic data is growing at an explosive rate.As a new method,deep learning not only has flexible structure design,but also can dig out the potential characteristics and rules of data from massive data.It has strong data learning ability and feature expression ability.Therefore,we apply deep learning methods to the field of traffic prediction,and conduct researches on traffic flow prediction in this thesis.The main research contents of this thesis are as follows:(1)Aiming at the traffic flow prediction of traffic section,a traffic flow prediction method based on attention mechanism and long short-term memory neural networks is proposed in this thesis.The method combines attention mechanism with long shortterm memory neural network.It uses long short-term memory neural network to extract the characteristics of the original traffic flow data,and then uses the attention mechanism to complete the compute of the attention weight of each feature,thereby achieving effective prediction of traffic flow.The proposed method not only can ensure the accuracy of traffic flow prediction,but also can help people better understand the temporal influence among the traffic flow conditions.(2)Aiming at the traffic flow prediction of traffic network,a traffic flow prediction method based on the generative adversarial network is proposed in this thesis.The method uses multi-layer long short-term memory neural networks to generate traffic flow data,and uses multi-layer fully connected layers to discriminate real traffic data and fake traffic data.This method is not only suitable for the prediction of traffic flow and speed,but also for single-step and multi-step prediction.Besides,the prediction accuracy of the proposed method is higher than the baseline models.