Traffic flow prediction method and device based on space-time residual network

Abstract

The invention discloses a traffic flow prediction method and device based on a space-time residual network. The method acquires geographic latitude and longitude of a to-be-detected city and the cityis mapped to an I*J grid according to the acquired geographic latitude and longitude at a preset ratio, wherein each square represents an area; the vehicle travel trajectory at each time period acquired by the GPS is counted, the acquired weather condition is recorded correspondingly, and collected data are obtained; the collected data are subjected to normalization preprocessing; the space-time residual network is trained according to the preprocessed data, and a trained traffic flow prediction model is formed; and the prediction model is called to predict the traffic flow in one area at a specific time and the prediction error is evaluated. When the method and the device are used for traffic flow prediction, through the traffic flow prediction model based on the space-time residual network, by using mutual influences among areas, the data can be mined and analyzed more deeply, and the traffic flow prediction accuracy is further improved to a great extent.

Description

technical field [0001] The invention relates to the technical field of intersecting deep learning and intelligent vehicle systems, in particular to a traffic flow prediction method and device based on a spatio-temporal residual network. Background technique [0002] With the development of society and the improvement of living standards, urban vehicles are increasingly congested, and vehicle accidents and air pollution are further aggravated. In order to predict vehicle congestion more accurately and provide more reasonable vehicle route planning for vehicle travel, it is necessary to carry out large-scale vehicle network congestion prediction. [0003] In the vehicle network, the traffic congestion status of the road is mutually influenced, and the traffic congestion status of a region is inseparable from the congestion status of adjacent regions, so the prediction of the dynamic changes of vehicles in each region needs to be considered from the global network; moreover, fo...

AI Technical Summary

Problems solved by technology

[0003] In the vehicle network, the traffic congestion status of the road is mutually influenced, and the traffic congestion status of a region is inseparable from the congestion status of adjacent regions, so the prediction of the dynamic changes of vehicles in each region needs to be considered from the global network; moreover, for a single Traffic congestion prediction on road sections is short-sighted. Local vehicle prediction is only based on historical data or based on the status of vehicles on limited road sections around. However, once the road section expands to large-scale vehicle network prediction, the amount of calculation will increase greatly, and the prediction efficiency and accuracy will be lower. Can not be guaranteed, can not meet the real-time and accuracy of car information service

[0004] The applicant of the present invention, after searching a large number of prior art documents, found that the existing vehicle prediction methods have the problems of low efficiency and low precision

Therefore, how to solve the problem that the existing methods cannot cross regions and store historical traffic flow data for a long time, so that they cannot predict the traffic flow more accurately according to the input and output traffic flow between adjacent regions, has become an important problem for those skilled in the art. research direction

Images