Titre: Simulation multi-agent de l’information des voyageurs dans les transports en commun.Résumé:Avec la généralisation de l'information temps-réel, le comportement des réseaux de transport modernes devient de plus en plus difficile à analyser et à prévoir. Le rôle de l'information est de plus en plus critique, particulièrement en cas de dysfonctionnement des réseaux, et l’information devient de plus en plus personnalisée et individuelle. Plusieurs phénomènes tels que la saturation, la concentration et la sur-réaction peuvent être observés après l’utilisation de systèmes d’information voyageurs. En effet, sans contrôle, la diffusion massive d'informations, à travers les panneaux à messages variables, les annonces dans les médias ainsi que les dispositifs de guidage individuel peut avoir des effets pervers et créer de nouvelles congestions. Ainsi, il est devenu important de développer des outils de simulation pour les décideurs de politiques de mobilité, prenant en compte ce nouvel environnement informationnel.Dans ce travail de thèse, nous proposons une simulation multi-agent pour mesurer l'impact de la fourniture d'informations sur la qualité des voyages en transports en commun, notamment dans des situations perturbées, en prenant en compte un environnement informationnel hétérogène. Dans un premier temps, nous concevons une simulation qui assure le déplacement de voyageurs sur un réseau de transport en commun. Ensuite, nous l’enrichissons par l’intégration de l’information des voyageurs et des taux d’équipements des voyageurs en smartphones, de telle manière qu’il puisse représenter les voyageurs connectés et être capable de distinguer l’impact des informations personnelles de celui des informations générales. Pour ce faire, nous nous fondons sur le paradigme multi-agent, qui est un modèle puissant pour la conception et l'implantation d'applications de transport. Pour répondre aux besoins de l’intégration de l’information des voyageurs, en particulier l’information individuelle, nous adoptons une approche centrée-environnement où l’environnement spatio-temporel multi-agent est l’interlocuteur des agents voyageurs et représente l’évolution dans le temps de l’état du réseau de transport en commun.Afin de tester notre simulateur dans un contexte réaliste de déplacement, nous utilisons les données réelles du réseau de Toulouse. Pour évaluer l’impact de la provision d’information voyageur sur le réseau, nous testons différents scénarios en fonction du pourcentage de voyageurs connectés représentés par des agents. Ces scénarios simulés sont analysés suivant leur impact sur les temps de parcours moyens des voyageurs, connectés et non connectés. Les résultats montrent que le nombre de voyageurs connectés a un impact positif sur les temps de parcours jusqu’à un certain seuil, au delà duquel l’impact devient relativement négatif / Title: Multiagent simulation of traveler information on transit networks.Abstract:With the generalization of real-time traveler information, the behavior of modern transport networks becomes harder to analyze and to predict. Advanced traveler Information systems play a major role in modern transportation system, mainly in case of disturbances, and the information is becoming more personalized and individual. Different phenomena such as over-saturation, concentration and over-reaction can be observed after the use of advanced traveler information systems. In fact, without control, the massive spread of information via billboards, radio announcements and individual guidance may have perverse effects and create new traffic jams. It is now critical to develop simulation tools for mobility policies makers, taking into account this new information environment to observe these effects and to consider the proper methods to deal with them.In this PHD work, we propose a multiagent simulation to measure the impact of information provision on the quality of passengers’ travels, notably in case of disturbances, taking into account a heterogeneous information environment. First, we design and implement a simulation to ensure travelers movement in a transit network. Then, we enrich our model to integrate traveler information system and to represent travelers equipped with smart phones. It allows us to evaluate separately personal and general information. To this end, we use the multi-agent paradigm, which is proven to be a powerful model to design and implement transportation applications. To deal with the integration of the traveler information system in the simulator, we adopt an environment-centered approach, where the space-time multiagent environment is the privileged interlocutor of the agents and represent the evolution of the transit network state over time.To test our simulator in a real context, we use real data on the city of Toulouse, France. To assess the impact of information provision, we simulate different scenarios in function of the percentage of connected travelers, represented as agents. These simulated scenarios are analyzed following their impact on the average travel times of the travelers (connected and no-connected). Results show that the number of connected travelers has a positive impact on overall travel times up until a certain threshold before becoming relatively negative

Les travaux conduits dans le cadre de cette thèse ont pour but d’étendre le champ de validité des simulations de trafic en milieu urbain et péri-urbain, avec notamment une meilleure prise en compte du contexte de conduite, de l’hétérogénéité des véhicules (deux-roues motorisés, véhicule léger, poids lourd, etc.) et des comportements des conducteurs en termes d’occupation de l’espace au sol. Le but est de produire en simulation des comportements observés en situation réelle tels que les phénomènes de faufilement (deux-roues, véhicules d’urgence, cas des intersections). Nous abordons, la simulation du trafic routier en considérant une approche comportementale fondée sur les systèmes multi-agents. Des travaux précédents ont proposé des solutions non génériques, notamment pour le cas particulier des deux-roues motorisés. Nous proposons un nouveau modèle d’agent permettant de mieux prendre en compte le contexte de conduite et les comportements des conducteurs en termes d’occupation de l’espace. Nous dotons l’agent d’une représentation ego-centrée de l’environnement fondée sur le concept d’affordance. Nous utilisons ainsi les affordances pour identifier les actions possibles, en termes d’occupation de l’espace, offertes par l’environnement. Nous les utilisons en- suite pour construire une représentation ego-centrée de la situation. Le modèle d’agent proposé a été implémenté avec Archisim. Cette implémentation a permis de reproduire une situation de trafic réelle et de comparer les données simulées avec celles recueillies sur le terrain. Différentes expérimentations ont été menées afin d’évaluer la qualité de la solution proposée. / The work described in this thesis aims to improve the validity of traffic simulations in urban and suburban areas. We propose to consider the driving context, the heterogeneity of vehicles (motorized two-wheelers, vehicle, truck, etc.) and the driver behavior in terms of space occupation. The purpose is to produce in simulation the observed real world phenomena such as faufilement behavior (two-wheelers, emergency vehicles, cases of intersections). We address the traffic simulation byconsidering a behavioral approach based on multi-agent systems. Previous work have proposed non- generic solutions, for the particular case of motorized two-wheelers. We propose a new agent model to better take into account the context of driving and driver behavior in terms of space occupation. We endow agent driver with an ego-centered representation of the environment based on the concept of affordances. We use the affordances to identify the possible actions, in terms of space occupation, afforded by environment. The proposed agent model was implemented with ArchiSim. This implementation has allowed to reproduce a real traffic situation and to compare the results to real data. Different experiments were conducted to assess the quality of the proposed solution.

Nous explorons le lien entre partage d'information (PI) et gestion collective de ressources naturelles (GRN). Pour déterminer ce lien, nous nous sommes interessé aux ostréiculteurs, acteurs sensibles à la qualité de l'eau et touchés par un virulent virus. Ces acteurs ont développé et utilisent divers artéfacts de PI. Ces artifacts sont destinés à faire face à des menaces potentielles. L'identification de ce point fondamental a mené à développer le concept de menaces. Elles sont définies par le modèle <A,C,I,D,E> (Acteurs, Caractéristiques, Infrastructure, Décisions, Environnement). Elles sont organisées le long de deux axes: l'internalité et l'exclusivité. Formuler la situation des ostréiculteurs en utilisant ce concept permet une caractérisation des enjeux pour les artéfacts de PI pour la lutte contre les menaces. Nous avons utilisé le cadre ENCORE pour une analyse qualitative et la SMA pour une quantitative de l’impact des artéfacts. La recherche montre qu'ils peuvent avoir des buts, media et contenus variés, améliorer la réflexivité, ou mener à peu, voire aucun changement. Ces améliorations sont liées au processus de création de l'artéfact. / I explore how information sharing (IS) and natural resources management (NRM) are linked.To determine this link, I focused on oyster farmers, actors sensitive to water quality and impacted by a virulent virus. Those actors implemented and use diverse IS artifacts. Those artifacts are meant to face potential threats. Realizing this focal point led to develop the threat concept. They are defined by the <A,C,I,D,E> (Actors, Characteristics, Infrastructure, Decisions, Environnement) model. They are organized along two axes: internality and excludability. Framing oyster farmers’ situation with this concept allow a characterization of stakes for IS artifacts to tackle threats.I used the ENCORE framework for qualitative assessment of IS artifacts impacts and a MAS for a quantitative one. The research shows that they have various goals, media and contents; can increase reflexivity or have little to no impact. Those changes are linked to artifact creation process.

Understanding how innovations are accepted in a dynamic and complex market environment is a crucial factor for competitive advantage. To understand the relevant factors for this diffusion and to predict success, empirically grounded agent-based models have become increasingly popular in recent years. Despite the popularity of these innovation diffusion models, no common framework that integrates their diversity exists. This article presents a flexible, modular and extensible common description and implementation framework that allows to depict the large variety of model components found in existing models. The framework aims to provide a theoretically grounded description and implementation framework for empirically grounded agent-based models of innovation diffusion. It identifies 30 component requirements to conceptualize an integrated formal framework description. Based on this formal description, a java-based implementation allowing for flexible configuration of existing and future models of innovation diffusion is developed. As a variable decision support tool in decision-making processes on the adoption of innovations the framework is valuable for the investigation of a range of research questions on innovation diffusion, business model evaluation and infrastructure transformation.

[pt] A ciência de dados precisa de uma grande quantidade de dados para testar e melhorar soluções. Jogos são largamente usados para abstrair situações da vida real. Rodadas de pôquer são um bom exemplo pois, por não saber as cartas dos oponentes, o jogador analisa um cenário de informação incompleta numa competição de agentes que envolve conhecimento probabilístico, análise de risco e brefe. Isso o diferencia de xadrez, damas e jogos de conhecimento perfeito e algoritmos de busca em forca bruta sobre o espaço de soluções. Usar o pôquer como um caso de teste possibilita a análise de diferentes abordagens usadas na vida real, porém num cenário mais controlado. Esta dissertação propõe um arcabouço de funcionalidades para criar e testar diferentes algorítimos de Deep Learning, que podem jogar pôquer entre sí, aprender com o histórico e maximizar suas recompensas. / [en] Data science research needs real examples to test and improve solutions. Games are widely used to mimic those real-world examples. Poker rounds are a good example of imperfect information state with competing agents dealing with probabilistic knowledge, risk assessment, and possible deception, unlike chess, checkers and perfect information brute-force search style of games. By using poker as a test-bed we can analyze different approaches used in real-world examples, in a more controlled environment, which should give great insights on how to tackle those real-world scenarios. We propose a framework to build and test different neural networks that can play against each other, learn from a supervised experience and maximize its rewards.

Submitted by Erika Demachki (erikademachki@gmail.com) on 2015-02-05T20:32:05Z No. of bitstreams: 3 Dissertação - Leonardo Afonso Amorim - 2014.pdf: 2774608 bytes, checksum: b212628d1bce8ef7bf3f80e7286db111 (MD5) Dissertação - Leonardo Afonso Amorim - 2014 - Projeto.zip: 11944741 bytes, checksum: 141e5b6b22a4f615ef5f5bee052d97b5 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Erika Demachki (erikademachki@gmail.com) on 2015-02-05T20:32:50Z (GMT) No. of bitstreams: 3 Dissertação - Leonardo Afonso Amorim - 2014.pdf: 2774608 bytes, checksum: b212628d1bce8ef7bf3f80e7286db111 (MD5) Dissertação - Leonardo Afonso Amorim - 2014 - Projeto.zip: 11944741 bytes, checksum: 141e5b6b22a4f615ef5f5bee052d97b5 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2015-02-05T20:32:50Z (GMT). No. of bitstreams: 3 Dissertação - Leonardo Afonso Amorim - 2014.pdf: 2774608 bytes, checksum: b212628d1bce8ef7bf3f80e7286db111 (MD5) Dissertação - Leonardo Afonso Amorim - 2014 - Projeto.zip: 11944741 bytes, checksum: 141e5b6b22a4f615ef5f5bee052d97b5 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2014-09-26 / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - CAPES / Decision making in public management is associated with a high degree of complexity due to insufficient financial resources to meet all the demands emanating from various sectors of society. Often, economic activities are in conflict with social or environmental causes. Another important aspect in decision making in public management is the inclusion of various stakeholders, eg public management experts, small business owners, shopkeepers, teachers, representatives of social and professional classes, citizens etc. The goal of this master thesis is to present two computational agents to aid decision making in public management as part of ADGEPA project: Miner Agent (MA) and Agent Decision Support (DSA). The MA uses data mining techniques and DSA uses multi-criteria analysis to point out relevant issues. The context in which this work fits is ADGEPA project. The ADGEPA (which means Digital Assistant for Participatory Public Management) is an innovative practice to support participatory decision making in public resources management. The main contribution of this master thesis is the ability to assist in the discovery of patterns and correlations between environmental aspects that are not too obvious and can vary from community to community. This contribution would help the public manager to make systemic decisions that in addition to attacking the main problem of a given region would decrease or solve other problems. The validation of the results depends on actual data and analysis of public managers. In this work, the data were simulated. / Tomada de decisão em gestão pública é associada ao alto grau de complexidade devido à insuficiência de recursos financeiros para atender todas as demandas provindas de diversos setores da sociedade. Frequentemente, atividades econômicas estão em conflito com causas sociais ou ambientais. Outro aspecto importante em tomadas de decisão em gestão pública é a inclusão dos diversos stakeholders, por exemplo especialistas em gestão pública, pequenos empresários, pequenos comerciantes, professores, representantes de classes sociais e profissionais, os próprios cidadãos etc. Diante disto, o objetivo deste trabalho de mestrado é apresentar uma proposta de Agente Minerador (AM) e Agente de Suporte à Decisão (ASD) para Gestão Pública Participativa e como fazer a interface entre eles. O AM faz uso de técnicas de mineração de dados para se encontrar regras de associação entre dados socioambientais, temporais e espaciais e o ASD faz uso de análise multicritério para ranquear problemas socioambientais que devem ser solucionados com prioridade. O contexto em que este trabalho se insere é o projeto ADGEPA (Assistente Digital para Gestão Pública Participativa), um projeto inovador para suporte à tomada de decisão participativa em gestão pública. Entende-se que a contribuição principal deste trabalho de mestrado é a possibilidade de auxiliar na descoberta de padrões e correlações entre aspectos socioambientais que não são muito óbvias e que podem variar de comunidade para comunidade. Esta contribuição poderá auxiliar o gestor público a tomar decisões sistêmicas que além de atacar o problema principal de uma determinada região diminuirá ou solucionará também problemas de outros aspectos. A validação dos resultados depende de dados reais e de análise de gestores públicos. Neste trabalho os dados foram simulados.