Traffic state prediction method and device

Abstract

The invention discloses a traffic state prediction method and device, and relates to the technical field of intelligent traffic. A specific embodiment of the method comprises the steps of determiningtraining sample data according to historical traffic data and first time period traffic data of a training road section and first time period traffic data of a road section adjacent to the training road section; wherein the historical traffic data is used as a first data source, the first time period traffic data of the training road section is used as a second data source, and the first time period traffic data of the road section adjacent to the training road section is used as a third data source; respectively determining the weight of each data source and the weight of each training sampleaccording to the training sample data; based on the training sample data, the weight of each data source, the weight of each training sample and a preset loss function, training to obtain a traffic model; and determining the traffic state of the to-be-predicted road section based on the traffic model. According to the embodiment, the influence of data disturbance can be effectively reduced, and the accuracy of traffic prediction is improved.

Description

technical field [0001] The invention relates to the technical field of intelligent transportation, in particular to a traffic state prediction method and device. Background technique [0002] In recent years, with the development of big data and machine learning technology, all walks of life have begun to apply machine learning technology to model and solve the problems of their corresponding scenarios. However, due to the instability of the underlying data source or noise disturbance, it is inevitable There will be a common problem: in the modeling and learning stage of the model, the data source and samples are optimized, which is a relatively "ideal" learning situation. However, after the model is actually launched, various data disturbance problems occur due to various reasons. As a result, the effect of the model after it goes online becomes worse, which leads to a series of subsequent problems. As a specific example, for example, in the case of traffic congestion, we ...

AI Technical Summary

Problems solved by technology

[0002] In recent years, with the development of big data and machine learning technology, all walks of life have begun to apply machine learning technology to model and solve the problems of their corresponding scenarios. However, due to the instability of the underlying data source or noise disturbance, it is inevitable There will be a common problem: in the modeling and learning stage of the model, the data source and samples are optimized, which is a relatively "ideal" learning situation. However, after the model is actually launched, various data disturbance problems occur due to various reasons. As a result, the effect of the model after it goes online becomes worse, which leads to a series of subsequent problems

[0004] In the process of realizing the present invention, the inventors found that there are at least the following problems in the prior art: (1) For the method of eliminating samples affected by the data disturbance problem during model prediction, although the accuracy of prediction can be guaranteed, when When the data disturbance is large, the scoring coverage of the model on the sample will drop greatly, resulting in unstable model coverage

For example, when a model uses data from 3 data sources, when a data source is missing as a whole, the sample coverage rate of this data source is 0; (2) Filling missing values ​​is a way to sacrifice prediction accuracy to reduce sample Loss, a method to improve the sample coverage rate. In the case of actual data disturbance, such an approach often leads to a large number of samples being concentrated, and it is difficult for the model to exert its due effect, that is, the actual online model effect is greatly reduced compared with the training time; (3) Training a tree model that is not sensitive to missing data has limitations in model selection, and when data sources or features are missing on a large scale, the tree model cannot solve the problem, resulting in a sharp drop in model effect

All in all, none of the commonly used methods can completely solve the problem of data perturbation and cannot obtain robust results.

Images