Large-scale traffic network jam prediction method and device based on convolutional neural network

Abstract

The invention discloses a large-scale traffic network jam prediction method and device based on a convolutional neural network. The time sequence and spatiality of road network vehicle speed information can be considered at the same time, so that the traffic jam state of the whole road network can be predicted more accurately. The method comprises the steps of S1, collecting GPS data of vehicles, extracting vehicle operating data at each moment on each road section, and generating a matrix M according to the acquired vehicle operating data; S2, generating a space-time thermodynamic diagram of at least one day according to the matrix M, wherein the abscissa of the space-time thermodynamic diagram represents time, and the ordinate of the space-time thermodynamic diagram represents road section ID sequence ranked according to spatial relationship; S3, generating a data set (X,Y) through window sliding on the space-time thermodynamic diagram; S4, establishing a convolutional neural network model, and training the convolutional neural network model by means of the data set (X,Y); S5, inputting a data set to be tested into the trained convolutional neural network model to obtain a prediction result.

Description

technical field [0001] The invention relates to the technical field of traffic information prediction, in particular to a large-scale traffic network congestion prediction method and device based on a convolutional neural network. Background technique [0002] In order to predict traffic congestion more accurately and provide more reasonable route planning for vehicle travel, it is necessary to carry out large-scale traffic network congestion prediction. Yes, the traffic congestion status of a region is inseparable from the congestion status of adjacent regions, so the prediction of traffic dynamic changes in each region needs to be based on the overall perspective of the network; (2) The traffic congestion prediction of a single road section is short-sighted, the most significant is the local Traffic forecasting only relies on historical data, or forecasts based on the traffic status of limited road sections around. When expanding from a single road section forecast to a la...

AI Technical Summary

Problems solved by technology

ARIMA can achieve certain results in the case of single-segment traffic speed prediction, but it cannot consider the entire network at the same time, especially the mutual influence between road segments, which limits the application of the model

(3) Modern machine learning algorithms, such as support vector machine SVM algorithm, SVM can generally achieve better results than regression by looking for high-dimensional separable or approximately separable planes of data, but algorithm training requires more time and computing memory , and cannot handle multi-output problems well, and cannot be applied at the large-scale network level; artificial neural network ANN is also applied to traffic prediction, and can predict multiple outputs at the same time, and can also be captured through continuous learning at the network layer There are certain patterns in the traffic data, but the ANN data-driven form does not understand the spatial relationship of the data well, and compared with the deep learning method, the ANN is too simple and the prediction accuracy is too low

[0004] The above technologies can better deal with the problem of local traffic congestion prediction, but when predicting the traffic speed of the whole network, the above models need to model the traffic speed information of each moment and each road section separately. When the span increases and the transportation network expands, the number of models to be constructed is extremely large; moreover, these large numbers of models exist independently, and the relationship of the entire transportation network cannot be considered at the same time

Images