Extract space-time dynamics from sensor network to build urban traffic prediction model: a machine learning point of view

Serge Fenet 1, 2 Yannick Perret 2 Julien Salotti 2, 3
1 STEEP - Sustainability transition, environment, economy and local policy
Inria Grenoble - Rhône-Alpes, LJK - Laboratoire Jean Kuntzmann
2 DM2L - Data Mining and Machine Learning
LIRIS - Laboratoire d'InfoRmatique en Image et Systèmes d'information
Abstract : Urban transport, while being essential for citizens to perform their daily activities, also constitutes one of the major sources of urban pollution (global emissions, local air quality, noise) and physical accidents (the two thirds of them taking place in cities), directly affecting humans health. Moreover, the economic impact of traffic jams on goods transportation within cities reduces the efficiency of urban delivery process and impair its’ economical aspect. In spite of great efforts made to favor modal report from car to other public transportation systems, the quest for an environmentally sustainable urban transport is a common and urgent challenge for all major cities in Europe. In these circumstances, and in order to (i) tackle the challenge of sustainable urban mobility, (ii) efficiently implement decision support tools, and (iii) exploit data allowing the assessment of policies and their resulting effects, urban planners need models. Along the years, different types of urban models have been developed, ranging from the static and aggregate land use-transportation interaction (LUTI) models, to more recent bottom-up, activity and agent-based simulation. In the context of the Optimod’Lyon project, this work focuses on building a traffic prediction model of road traffic within the city of Lyon by using the Grand Lyon historical measures from more than 650 sensors between the year 2007 and now. This global model will be composed of a predictive model (gray box model that preserves dynamics) and an explanatory model (white box model that explicitly identify space-time features). This paper aims at presenting the ongoing process of exploration of these data and the building of these models. Each section of this abstract highlight a potential barrier, and quickly presents the possible tracks a modeller could use to overcome them. The main conclusion we wish to convey is that, as almost every question is a research field in itself, important choices have to be made at every step, and there can be no universal model.
Type de document :
Communication dans un congrès
Urban Modelling Symposium, Oct 2014, Lyon, France. 2014
Liste complète des métadonnées

https://hal.archives-ouvertes.fr/hal-01241516
Contributeur : Serge Fenet <>
Soumis le : jeudi 10 décembre 2015 - 15:54:41
Dernière modification le : lundi 16 juillet 2018 - 15:10:31
Document(s) archivé(s) le : samedi 29 avril 2017 - 11:11:52

Fichier

urban_modelling_Fenet_et_al_20...
Fichiers produits par l'(les) auteur(s)

Identifiants

  • HAL Id : hal-01241516, version 1

Citation

Serge Fenet, Yannick Perret, Julien Salotti. Extract space-time dynamics from sensor network to build urban traffic prediction model: a machine learning point of view. Urban Modelling Symposium, Oct 2014, Lyon, France. 2014. 〈hal-01241516〉

Partager

Métriques

Consultations de la notice

914

Téléchargements de fichiers

148