Les travaux de cette thèse visent à mettre en place une gestion prédictive sous incertitudes de la ressource en eau pour les réseaux de voies navigables. L'objectif est de proposer un plan de gestion de l'eau pour optimiser les conditions de navigation de l'ensemble du réseau supervisé sur un horizon spécifié. La solution attendue doit rendre le réseau résilient aux effets probables du changement climatique et aux évolutions du trafic fluvial. Dans un premier temps, une modélisation générique d'une ressource distribuée sur un réseau est proposée. Celle-ci, basée sur les processus décisionnels markoviens, prend en compte les nombreuses incertitudes affectant les réseaux considérés. L'objectif de cette modélisation est de couvrir l'ensemble des cas possibles, prévus ou non, afin d'avoir une gestion résiliente de ces réseaux. La seconde contribution consiste en une distribution du modèle sur plusieurs agents afin de permettre son passage à l'échelle. Ceci consiste en une répartition des capacités de contrôle du réseau entre les agents. Chaque agent ne possède ainsi qu'une connaissance locale du réseau supervisé. De ce fait, les agents ont besoin de se cordonner pour proposer une gestion efficace du réseau. Une résolution itérative avec échanges de plans temporaires de chaque agent est utilisée pour l'obtention de politiques de gestion locales à chaque agent. Finalement, des expérimentations ont été réalisées sur des réseaux réels de voies navigables françaises pour observer la qualité des solutions produites. Plusieurs scénarios climatiques différents ont été simulés pour tester la résilience des politiques produites. / The work of this thesis aims to introduce and implement a predictive management under uncertainties of the water resource for inland waterway networks. The objective is to provide a water management plan to optimize the navigation conditions of the entire supervised network over a specified horizon. The expected solution must render the network resilient to probable effects of the climate change and changes in waterway traffic. Firstly, a generic modeling of a resource distributed on a network is proposed. This modeling, based on Markovian Decision Processes, takes into account the numerous uncertainties affecting considered networks. The objective of this modeling is to cover all possible cases, foreseen or not, in order to have a resilient management of those networks. The second contribution consists in a distribution of the model over several agents to facilitate the scaling. This consists of a repartition of the network's control capacities among the agents. Thus, each agent has only local knowledge of the supervised network. As a result, agents require coordination to provide an efficient management of the network. An iterative resolution, with exchanges of temporary plans from each agent, is used to obtain local management policies for each agent. Finally, experiments were carried out on realistic and real networks of the French waterways to observe the quality of the solutions produced. Several different climatic scenarios have been simulated to test the resilience of the produced policies.

La présente thèse de doctorat en informatique propose un modèle pour une recherche d'information intelligente possibiliste des documents Web et son implémentation. Ce modèle est à base de deux Réseaux Petits Mondes Hiérarchiques (RPMH) et d'un Réseau Possibiliste (RP) : Le premier RPMH consiste à structurer les documents retrouvés en zones denses de pages Web thématiquement liées les unes aux autres. Nous faisons ainsi apparaître des nuages denses de pages qui traitent d'un sujet et des sujets connexes (assez similaires sémantiquement) et qui répondent toutes fortement à une requête. Le second RPMH est celui qui consiste à ne pas prendre les mots-clés tels quels mais à considérer une requête comme multiple en ce sens qu'on ne cherche pas seulement le mot-clé dans les pages Web mais aussi les substantifs qui lui sont sémantiquement proches. Les Réseaux Possibilistes combinent les deux RPMH afin d'organiser les documents recherchés selon les préférences de l'utilisateur. En effet, l'originalité du modèle proposé se décline selon les trois volets suivants qui synthétisent nos contributions. Le premier volet s'intéresse au processus itératif de la reformulation sémantique de requêtes. Cette technique est à base de relations de dépendance entre les termes de la requête. Nous évaluons notamment les proximités des mots du dictionnaire français « Le Grand Robert » par rapport aux termes de la requête. Ces proximités sont calculées par le biais de notre approche de recherche des composantes de sens dans un RPMH de dictionnaire de mots par application d'une méthode basée sur le dénombrement des circuits dans le réseau. En fait, l'utilisateur du système proposé choisit le nombre de mots sémantiquement proches qu'il désire ajouter à chaque terme de sa requête originelle pour construire sa requête reformulée sémantiquement. Cette dernière représente la première partie de son profil qu'il propose au système. La seconde partie de son profil est constituée des choix des coefficients de pertinence possibilistes affectés aux entités logiques des documents de la collection. Ainsi, notre système tient compte des profils dynamiques des utilisateurs au fur et à mesure que ces derniers utilisent le système. Ce dernier est caractérisé par son intelligence, son adaptativité, sa flexibilité et sa dynamicité. Le second volet consiste à proposer des relations de dépendance entre les documents recherchés dans un cadre ordinal. Ces relations de dépendance entre ces documents traduisent les liens sémantiques ou statistiques évaluant les distributions des termes communs à des paires ou ensembles de documents. Afin de quantifier ces relations, nous nous sommes basés sur les calculs des proximités entres ces documents par application d'une méthode de dénombrement de circuits dans le RPMH de pages Web. En effet, les documents peuvent ainsi être regroupés dans des classes communes (groupes de documents thématiquement proches). Le troisième volet concerne la définition des relations de dépendance, entre les termes de la requête et les documents recherchés, dans un cadre qualitatif. Les valeurs affectées à ces relations traduisent des ordres partiels de préférence. En fait, la théorie des possibilités offre deux cadres de travail : le cadre qualitatif ou ordinal et le cadre quantitatif. Nous avons proposé notre modèle dans un cadre ordinal. Ainsi, des préférences entre les termes de la requête se sont ajoutées à notre modèle de base. Ces préférences permettent de restituer des documents classés par préférence de pertinence. Nous avons mesuré aussi l'apport de ces facteurs de préférence dans l'augmentation des scores de pertinence des documents contenant ces termes dans le but de pénaliser les scores de pertinence des documents ne les contenant pas. Pour la mise en place de ce modèle nous avons choisi les systèmes multi-agents. L'avantage de l'architecture que nous proposons est qu'elle offre un cadre pour une collaboration entre les différents acteurs et la mise en œuvre de toutes les fonctionnalités du système de recherche d'information (SRI). L'architecture s'accorde parfaitement avec le caractère intelligent possibiliste et permet de bénéficier des capacités de synergie inhérente entre les différentes composantes du modèle proposé. Dans le présent travail, nous avons donc pu mettre en exergue à travers les expérimentations effectuées l'intérêt de faire combiner les deux RPMH via un réseau possibiliste dans un SRI, ce qui permet d'enrichir le niveau d'exploration d'une collection. Ce dernier n'est pas limité aux documents mais l'étend en considérant les requêtes. En effet, la phase de reformulation sémantique de requête permet à l'utilisateur de profiter des autres documents correspondants aux termes sémantiquement proches des termes de la requête originelle. Ces documents peuvent exister dans d'autres classes des thèmes. En conséquence, une reclassification proposée par le système s'avère pertinente afin d'adapter les résultats d'une requête aux nouveaux besoins des utilisateurs. / This Ph.D. thesis proposes a new model for a multiagent possibilistic Web information retrieval and its implementation. This model is based on two Hierarchical Small-Worlds (HSW) Networks and a Possibilistic Networks (PN): The first HSW consists in structuring the founded documents in dense zones of Web pages which strongly depend on each other. We thus reveal dense clouds of pages which "speak" more or less about the same subject and related subjects (semantically similar) and which all strongly answer user's query. The second HSW consists in considering the query as multiple in the sense that we don't seek only the keyword in the Web pages but also its semantically close substantives. The PN generates the mixing of these two HSW in order to organize the searched documents according to user's preferences. Indeed, the originality of the suggested model is declined according to three following shutters' which synthesize our contributions. The first shutter is interested in the iterative process of query semantic reformulation. This technique is based on relationship dependence between query's terms. We evaluate in particular the semantics proximities between the words of the French dictionary "Le Grand Robert" and query's terms. These proximities are calculated via our approach of research of the semantics components in the HSW of dictionary of words by application of our method of enumeration of circuits in the HSW of dictionary. In fact, the user of the suggested system chooses the number of close words that he desire to add to each word of his initial query to build his semantically reformulated query. This one represents the first part of user's profile which he proposes to the system. The second part of its profile makes up of its choices of the coefficients of relevance possibilistic of the logical entities of the documents of the collection. Thus, our system takes account of the dynamic profiles of its users progressively they use the system, which proves its intelligence, its adaptability, its flexibility and its dynamicity. The second shutter consists in proposing relationship dependence between documents of the collection within an ordinal framework. These relationships dependence between these documents represent the semantic or statistical links evaluating the distributions of the general terms to pairs or sets of documents.  In order to quantify these relationships, we are based on the calculations of the proximities between these documents by application of a method enumerating of circuits in the HSW of Web pages. Indeed, the documents can thus be clustered in common classes (groups of close documents). The third shutter is related to the definition of the relationships dependence between query's terms and documents of the collection, within a qualitative framework. The assigned values to these relations translate preferably partial orders. In fact, possibilistic theory offers two working frameworks:  the qualitative or ordinal framework and the numerical framework.  We proposed our model within an ordinal framework. Thus, we add to our basic model preferences between query's terms. These preferences make it possible to restore documents classified by relevance's preference. We also measured the contribution of these preferably factors in the increase of the relevance's scores of  documents containing these terms with an aim of penalizing the relevance's scores of the documents not containing them. For the installation of this model we chose multiagent systems. The advantage of the proposed architecture is that it offers a framework for collaboration between the various actors and the implementation of all the functionalities of the information retrieval system. Architecture agrees perfectly with the possibilistic intelligent character and makes it possible to profit from the capacities of inherent synergy in the suggested model. We thus could put forward, through the carried out experiments, the goal of combining the two HSW via a possibilistic network in an information retrieval system, which makes it possible to enrich the exploration level of a collection. This exploration is not only limited to the documents but it extends by considering also the query. Indeed, the semantic query reformulation phase makes it possible to benefit user from other documents which contain some close terms of the initial query. These documents can exist in other topics classes. Consequently, a reclassification suggested by the system proves its relevance in order to adapt query's results to new user's needs.

For researches in the field of the data mining and machine learning the necessary condition is an availability of various input data set. Now researchers create the databases of such sets. Examples of the following systems are: The UCI Machine Learning Repository, Data Envelopment Analysis Dataset Repository, XMLData Repository, Frequent Itemset Mining Dataset Repository. Along with above specified statistical repositories, the whole pleiad from simple filestores to specialized repositories can be used by researchers during solution of applied tasks, researches of own algorithms and scientific problems. It would seem, a single complexity for the user will be search and direct understanding of structure of so separated storages of the information. However detailed research of such repositories leads us to comprehension of deeper problems existing in usage of data. In particular a complete mismatch and rigidity of data files structure with SDMX - Statistical Data and Metadata Exchange - standard and structure used by many European organizations, impossibility of preliminary data origination to the concrete applied task, lack of data usage history for those or other scientific and applied tasks. Now there are lots of methods of data miming, as well as quantities of data stored in various repositories. In repositories there are no methods of DM (data miming) and moreover, methods are not linked to application areas. An essential problem is subject domain link (problem domain), methods of DM and datasets for an appropriate method. Therefore in this work we consider the building problem of ontological models of DM methods, interaction description of methods of data corresponding to them from repositories and intelligent agents allowing the statistical repository user to choose the appropriate method and data corresponding to the solved task. In this work the system structure is offered, the intelligent search agent on ontological model of DM methods considering the personal inquiries of the user is realized. For implementation of an intelligent data and metadata exchange repository the agent oriented approach has been selected. The model uses the service oriented architecture. Here is used the cross platform programming language Java, multi-agent platform Jadex, database server Oracle Spatial 10g, and also the development environment for ontological models - Protégé Version 3.4.

Thesis aims at fighter combat and maneuvring - so called Dogfighting. The purpose of this work is to create intelligent autopilot based on Agent system, eligible of executing in-air maneuvers and tactics in real-time simulation. In the first part, theoretical basis of air combat will be introduced, such as weapon systems, maneuvring and tactics in mutual combat 1 on 1, odds fight 2 on 1, and last but not least mass fights. Also agent programming will be introduced, as well as recognizing of agent rules and processes and its transformation to agent language. The second part describes building of a simple graphical simulation environment based on JMonkey game engine. Agent system maintaining every single aircraft within the simulation will be created and own network socket protocol for communication between intelligent behavior and simulation environment will be discussed and documented.

No description available.

Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / The goal of this work is to present MP-Draughts (MultiPhase- Draughts), that is a multiagent environment for Draughts, where one agent - named IIGA- is built and trained such as to be specialized for the initial and the intermediate phases of the games and the remaining ones for the final phases of them. Each agent of MP-Draughts is a neural network which learns almost without human supervision (distinctly from the world champion agent Chinook). MP-Draughts issues from a continuous activity of research whose previous product was the efficient agent VisionDraughts. Despite its good general performance, VisionDraughts frequently does not succeed in final phases of a game, even being in advantageous situation compared to its opponent (for instance, getting into endgame loops). In order to try to reduce this misbehavior of the agent during endgames, MP-Draughts counts on 25 agents specialized for endgame phases, each one trained such as to be able to deal with a determined cluster of endgame boardstates. These 25 clusters are mined by a Kohonen-SOM Network from a Data Base containing a large quantity of endgame boardstates. After trained, MP-Draughts operates in the following way: first, an optimized version of VisionDraughts is used as IIGA; next, the endgame agent that represents the cluster which better fits the current endgame board-state will replace it up to the end of the game. This work shows that such a strategy significantly improves the general performance of the player agents. / O objetivo deste trabalho é propor um sistema de aprendizagem de Damas, o MPDraughts (MultiPhase- Draughts): um sistema multiagentes, em que um deles - conhecido como IIGA (Initial/Intermediate Game Agent)- é desenvolvido e treinado para ser especializado em fases iniciais e intermediárias de jogo e os outros 25 agentes, em fases finais. Cada um dos agentes que compõe o MP-Draughts é uma rede neural que aprende a jogar com o mínimo possível de intervenção humana (distintamente do agente campeão do mundo Chinook). O MP-Draughts é fruto de uma contínua atividade de pesquisa que teve como produto anterior o VisionDraughts. Apesar de sua eficiência geral, o Vision- Draughts, muitas vezes, tem seu bom desempenho comprometido na fase de finalização de partidas, mesmo estando em vantagem no jogo em comparação com o seu oponente (por exemplo, entrando em loop de final de jogo). No sentido de reduzir o comportamento indesejado do jogador, o MP-Draughts conta com 25 agentes especializados em final de jogo, sendo que cada um é treinado para lidar com um determinado tipo de cluster de tabuleiros de final de jogo. Esses 25 clusters são minerados por redes de Kohonen-SOM de uma base de dados que contém uma grande quantidade de estado de tabuleiro de final de jogo. Depois de treinado, o MP-Draughts atua da seguinte maneira: primeiro, uma versão aprimorada do VisionDraughts é usada como o IIGA; depois, um agente de final de jogo que representa o cluster que mais se aproxima do estado corrente do tabuleiro do jogo deverá substituir o IIGA e conduzir o jogo até o final. Este trabalho mostra que essa estratégia melhorou, significativamente, o desempenho geral do agente jogador. / Mestre em Ciência da Computação