Prédiction de séries temporelles multi-variées stationnaires: modélisation du contexte pour l'analyse des données de transports

Abstract : This article presents a new predictive model for multivariate time series that not only makes use of the relations between past and future values but also learns the representation of some exogenous features such as spatio-temporal context. We demonstrate the advantages of our model by applying it to the forecast of smart cards tap-in logs in the Parisian subway network: our model (that uses Recurent Neural Network to learn representations) outperforms the baseline for forecasting while also elegantly learning the spatio-temporal context.
Complete list of metadatas

Cited literature [2 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02269214
Contributor : Perrine Cribier-Delande <>
Submitted on : Thursday, August 22, 2019 - 5:27:54 PM
Last modification on : Thursday, September 5, 2019 - 1:26:25 AM

Files

ArticleGRETSI2019.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02269214, version 1

Citation

Valentin Guiguet, Perrine Cribier-Delande, Nicolas Baskiotis, Vincent Guigue. Prédiction de séries temporelles multi-variées stationnaires: modélisation du contexte pour l'analyse des données de transports. GRETSI 2019, Aug 2019, Lille, France. ⟨hal-02269214⟩

Share

Metrics

Record views

37

Files downloads

15