Entrepôts, Représentation et Ingénierie des Connaissances
Publications du laboratoire

Recherche approfondie

par Année
par Auteur
par Thème
par Type
- Short-term urban rail passenger flow forecasting: a dynamic bayesian network approach

Auteur(s): Roos J.(Corresp.), Bonnevay S., Gavin G.

Conference: International Conference on Machine Learning and Applications (Anaheim, US, 2016-12-18)
Actes de conférence: , vol. p.1034-1039 (2016)


In this paper, we propose a dynamic Bayesian net- work approach to forecast the short-term passenger flows of the urban rail network of Paris. The structure of the model is based on the dependencies between each flow and its upstream flows at previous time slices. The conditional probability distributions are described as linear Gaussians. We carry out the experiment on an entire metro line, using ticket validation, count and transport supply data. Given the presence of incomplete data, we perform a structural expectation-maximization (SEM) algorithm both to learn the structure and to find the maximum likelihood estimate of the parameters. Finally, short-term forecasting is conducted by inference using the bootstrap filter. One of the main advantages of this model is its ability to forecast in the presence of incomplete data. For most passenger flows, the forecasting results outperform naïve methods such as historical average and last observation carried forward (LOCF). These results illustrate the potential of the approach, as well as the fundamental role of the transport supply data in the modeling.