Vous êtes ici : GIPSA-lab > Formation > Thèses soutenues

'Control of large traffic networks'


Directeur de thèse :     Carlos CANUDAS-DE-WIT

Co-encadrant :     Federica GARIN

École doctorale : Electronique, electrotechnique, automatique, traitement du signal (EEATS)

Spécialité : Automatique et productique

Structure de rattachement : CNRS

Établissement d'origine : Université de Calabre- Italie

Financement(s) : Contrat doctoral


Date d'entrée en thèse : 23/04/2014

Date de soutenance : 11/09/2017


Composition du jury :
M. CANUDAS DE WIT Carlos, Directeur de thèse
Mme SACONE Simona, Professeur associé à l'Université de Gênes, Italie, rapporteur
M. CAMACHO Eduardo, Professeur à l''Université de Séville, Espagne, rapporteur
M. ALAMIR Mazen, Directeur de recherche au CNRS, examinateur et président
Mme GOATIN Paola, Directeur de recherche à Inria, examinateur
M. DE NUNZIO Giovanni, Ingénieur de recherhce à l''IFP-EN, examinateur
Mme GARIN Federica, Chargée de recherche à Inria, examinateur


Résumé : The thesis focuses on traffic lights control in large scale urban networks. It starts off with a study of macroscopic modeling based on the Cell Transmission model. We formulate a signalized version of such a model in order to include traffic lights' description into the dynamics. Moreover, we introduce two simplifications of the signalized model towards control design, one that is based on the average theory and considers duty cycles of traffic lights, and a second one that describes traffic lights trajectories with the time instants of the rising and falling edges of a binary signals. We use numerical simulations to validate the models with respect to the signalized Cell Transmission model, and microsimulations (with the software Aimsun), to validate the same model with respect to realistic vehicles' behavior.
We propose two control algorithms based on the two models mentioned above. The first one, that uses the average Cell Transmission model, considers traffic lights' duty cycles as controlled variables and it is formulated as an optimization problem of standard traffic measures. We analyze such a problem and we show that it is equivalent to a convex optimization problem, so ensuring its computational efficiency. We analyze its performance with respect to a best-practice control scheme both in MatLab simulations and in Aimsun simulations that emulate a large portion of Grenoble, France. The second proposed approach is an optimization problem in which the decision variables are the activation and deactivation time instants of every traffic lights. We employ the Big-M modeling technique to reformulate such a problem as a mixed integer linear program, and we show via numerical simulations that the expressivity of it can lead to improvements of the traffic dynamics, at the price of the computational efficiency of the control scheme.
To pursue the scalability of the proposed control techniques we develop two iterative distributed approaches to the traffic lights control problem. The first, based on the convex optimization above mentioned, uses the dual decomposition technique and is provably optimal, that is, it gives the same solution as the centralized optimization. The second, based on the mixed integer problem aforesaid, is a suboptimal algorithm that leads to substantial improvements by means of the computational efficiency with respect to the related centralized problem. We analyze via numerical simulations the convergence speed of the iterative algorithms, their computational burden and their performance regarding traffic metrics.
The thesis is concluded with a study of the traffic lights control algorithm that is employed in several large intersections in Grenoble. We present the working principle of such an algorithm, detailing technological and methodological differences with our proposed approaches. We create into Aimsun the scenario representing the related part of the city, also reproducing the control algorithm and comparing its performance with the ones given by one of our approaches on the same scenario.

GIPSA-lab, 11 rue des Mathématiques, Grenoble Campus BP46, F-38402 SAINT MARTIN D'HERES CEDEX - 33 (0)4 76 82 71 31