Méthodes de sélection de voisinage et de prévision à court-terme pour l’analyse du trafic urbain

Abstract : In the context of Smart Cities, the need to inform drivers, to anticipate and to take action to regulate the traffic flow becomes critical. This need has driven the development of a large number of short-term (less than one hour) traffic flow forecasting methods. The era of big data has seen the rise in computing power, in storage capacity and in our ability to process information in real-time. At the same time, more and more road segments are equipped with traffic sensors. This evolution of technology is reflected in the evolution of traffic forecasting methods. In this work, we explore multiple questions in order to improve the accuracy of forecasting models on traffic data. The first question deals with the spatio-temporal neighborhood : what information should we consider in order to predict the future activity on a sensor? The second question is about the choice of the best forecasting methods with respect to the nature of the network (urban, freeway) and the forecast horizon. The last question concerns the choice of the optimal temporal aggregation for the data and its impact on the forecasting accuracy. In order to investigate these questions, we have studied a large panel of forecasting methods (ARIMA, VAR, k-NN, SVR, neural networks) and two variable selection mechanisms (Lasso, TiGraMITe). This experimental study has been conducted on data from the urban network of Lyon and from urban freeway of Marseille.
Document type :
Theses
Complete list of metadatas

Cited literature [129 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/tel-02339366
Contributor : Julien Salotti <>
Submitted on : Thursday, November 7, 2019 - 4:57:49 PM
Last modification on : Tuesday, November 19, 2019 - 2:44:01 AM

File

J.Salotti_Manuscrit_These.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : tel-02339366, version 1

Citation

Julien Salotti. Méthodes de sélection de voisinage et de prévision à court-terme pour l’analyse du trafic urbain. Apprentissage [cs.LG]. INSA Lyon; Université de Lyon, 2019. Français. ⟨tel-02339366⟩

Share

Metrics

Record views

68

Files downloads

38