Research And Impletation Of Traffic Speed Prediction Algorithm Based On Deep Learning

With the rapid development of economy and the continuous acceleration of urban construction,the urban road network has been improving.Correspondingly,the vehicle ownership of urban households has increased significantly,and the road congestion during peak hours is not optimistic.Therefore,how to coordinate and improve the traffic conditions of the entire city so as to make the best use of traffic resources has become an urgent problem.Building an intelligent traffic assistance system is obviously a good solution.For an intelligent transportation assistance system,the city-scale road speed prediction provides important data support,which can not only provide a reference for the management and decision-making of the transportation department,but also give a reference to the urban road planning department.This design has important practical significance for ensuring the smooth operation of urban road traffic.In recent years,due to the increasing number of sensors deployed on urban roads and the fusion of various data,the amount of historical and real-time traffic data has increased rapidly,which providing a good data environment for road speed prediction.Accurate prediction of road speeds can help us to know the overall traffic situation of urban roads,which plays an important role in many ways.For the traffic management department,this helps them formulate scheduling plans in advance,scientifically manage traffic flow,and make the best use of urban traffic resources,thereby reducing traffic jams and even traffic accidents.For the driver,this can provide them with an actual estimate of the journey time,which will reduce uncertainty during the journey.In addition,the prediction of traffic conditions is helpful for route selection,alternative route recommendations,and expected delay assessment of navigation applications.In short,whether it is for traffic management departments or drivers and navigation applications,it is of great value to accurately predict the urban road traffic conditions,and it is obviously more reasonable to make decisions based on the prediction results than just based on the current traffic information.However,it is very challenging to accurately predict the road speed of the entire city,because the traffic condition of urban roads may be affected by various factors.It can be divided into three categories:time,space and other potential information,such as abnormal events.Although the road speed prediction algorithm has developed rapidly and made some achievements with the help of related big data technology,it still faces multiple challenges.On the one hand,many existing methods treat traffic prediction only as a time series problem,but the traffic condition of a road segment is strongly related to the situation of other road segments,so the overall information of the entire urban road traffic network must not be ignored;on the other hand,some methods introduces additional spatiotemporal data to assist traffic prediction.Although it solves the problem of global information to a certain extent,the spatiotemporal correlation of city road system shows that traffic prediction is not equal to global prediction using local information,and additional information will consume a lot of computation.Aiming at the problems of the current road speed prediction algorithm,this paper proposes a road speed prediction algorithm,L-U-Net,based on spatiotemporal features.The algorithm is a spatiotemporal prediction model based on convolutional neural network and long-short-term memory network,which can achieve accurate and effective prediction of spatiotemporal problems.This algorithm is a new spatiotemporal prediction model based on convolutional neural network and short-term memory network,which can realize accurate and effective prediction of spatiotemporal problems.Compared with other algorithms,it has obvious advantages in workload and complexity of feature engineering.More importantly,it can avoid the high complexity and uncertainty of subjective feature extraction.Experiments show that the L-U-Net algorithm can accurately and effectively predict the future speed of urban roads,which provides important ideas for the study of prediction problems with spatial and temporal relationships.Next,for the impact of some abnormal events on the speed of urban roads,a road speed prediction algorithm based on abnormal events is proposed in this paper,which is TCPM.The algorithm is based on convolutional neural network for modeling,and it can predict the road speed of the entire city for certain abnormal events that may occur in the future.Its characteristic is that in addition to considering the time and space information,it emphasizes other potential information,that is,the influence of abnormal events on the road speed prediction.Experiments prove that the TCPM algorithm can effectively predict the future speed of urban roads and also shows that the use of convolutional neural networks to extract feature vectors can optimize the results of speed prediction,which has guiding significances for research in related fields.Finally,based on the above two urban road speed prediction algorithms,this paper designs and implements a road speed prediction algorithm display system,which integrates history data display,model display,speed prediction display and other functions.All these intuitively show the research results of this paper.