Nous considérons une plate-forme pair-à-pair pour la Recherche d'Information (RI) collaborative. Chaque pair héberge une collection de documents textuels qui traitent de ses sujets d'intérêt. En l'absence d'un mécanisme d'indexation global, les pairs indexent localement leurs documents et s'associent pour fournir un service distribué de réponse à des requêtes. Notre objectif est de concevoir un protocole décentralisé qui permette aux pairs de collaborer afin de transmettre une requête depuis son émetteur jusqu'aux pairs en possession de documents pertinents. Les réseaux logiques sémantiques (Semantic Overlay Networks, SON) représentent la solution de référence de l'état de l'art. Les pairs qui possèdent des ressources sémantiques similaires sont regroupés en clusters. Les opérations de RI seront alors efficaces puisqu'une requête sera transmise aux clusters de pairs qui hébergent les ressources pertinentes. La plupart des approches actuelles consistent en une reconfiguration dynamique du réseau de pairs (peer rewiring). Pour ce faire, chaque pair exécute périodiquement un algorithme de marche aléatoire ou gloutonne sur le réseau pair-à-pair afin de renouveler les pairs de son cluster. Ainsi, un réseau à la structure initialement aléatoire évolue progressivement vers un réseau logique sémantique. Jusqu'à présent, les approches existantes n'ont pas considéré que l'évolution de la topologie du réseau puisse influer sur les performances de l'algorithme de reconfiguration dynamique du réseau. Cependant, s'il est vrai que, pour une configuration initiale aléatoire des pairs, une marche aléatoire sera efficace pour découvrir les pairs similaires, lorsque des clusters commencent à émerger une approche gloutonne devient alors mieux adaptée. Ainsi, nous proposons une stratégie qui applique un algorithme de recuit simulé (Simulated Annealing, SA) afin de faire évoluer une stratégie de marche aléatoire vers une stratégie gloutonne lors de la construction du SON. Cette thèse contient plusieurs avancées concernant l'état de l'art dans ce domaine. D'abbord, nous modélisions formellement la reconfiguration dynamique d'un réseau en un SON. Nous identifions un schéma générique pour la reconfiguration d'un réseau pair-à-pair, et après le formalisons en une procédure constituée de trois étapes. Ce framework cohérent offre à ses utilisateurs de quoi le paramétrer. Ensuite, le problème de la construction d'un SON est modélisé sous la forme d'un problème d'optimisation combinatoire pour lequel les opérations de reconfiguration du réseau correspondent à la recherche décentralisée d'une solution locale. Fondée sur ce modèle, une solution concrète à base de recuit simulé est proposée. Nous menons une étude expérimentale poussée sur la construction du SON et la RI sur SONs, et validions notre approche. / A Peer-to-Peer (P2P) platform is considered for collaborative Information Retrieval (IR). Each peer hosts a collection of text documents with subjects related to its owner's interests. Without a global indexing mechanism, peers locally index their documents, and provide the service to answer queries. A decentralized protocol is designed, enabling the peers to collaboratively forward queries from the initiator to the peers with relevant documents. Semantic Overlay Network (SONs) is one the state of the art solutions, where peers with semantically similar resources are clustered. IR is efficiently performed by forwarding queries to the relevant peer clusters in an informed way. SONs are built and maintained mainly via peer rewiring. Specifically, each peer periodically sends walkers to its neighborhood. The walkers walk along peer connections, aiming at discovering more similar peers to replace less similar neighbors of its initiator. The P2P network then gradually evolves from a random overlay network to a SON. Random and greedy walk can be applied individually or integrated in peer rewiring as a constant strategy during the progress of network evolution. However, the evolution of the network topology may affect their performance. For example, when peers are randomly connected with each other, random walk performs better than greedy walk for exploring similar peers. But as peer clusters gradually emerge in the network, a walker can explore more similar peers by following a greedy strategy. This thesis proposes an evolving walking strategy based on Simulated Annealing (SA), which evolves from a random walk to a greedy walk along the progress of network evolution. According to the simulation results, SA-based strategy outperforms current approaches, both in the efficiency to build a SON and the effectiveness of the subsequent IR. This thesis contains several advancements with respect to the state of the art in this field. First of all, we identify a generic peer rewiring pattern and formalize it as a three-step procedure. Our technique provides a consistent framework for peer rewiring, while allowing enough flexibility for the users/designers to specify its properties. Secondly, we formalize SON construction as a combinatorial optimization problem, with peer rewiring as its decentralized local search solution. Based on this model, we propose a novel SA-based approach to peer rewiring. Our approach is validated via an extensive experimental study on the effect of network wiring on (1) SON building and (2) IR in SONs.

This dissertation studies models for locating facilities in time varying demand environments. We describe the characteristics of the time varying demand that motivate the analysis of our location models in terms of total demand and the change in value and location of the demand of each customer. The first part of the dissertation is devoted to the dynamic location model, which determines the optimal time and location for establishing capacitated facilities when demand and cost parameters are time varying. This model minimizes the total cost over a discrete and finite time horizon for establishing, operating, and closing facilities, including the transportation costs for shipping demand from facilities to customers. The model is solved using Lagrangian relaxation and Benders? decomposition. Computational results from different time varying total demand structures demonstrate, empirically, the performance of these solution methods. The second part of the dissertation studies two location models where relocation of facilities is not allowed and the objective is to determine the optimal location of capacitated facilities that will have a good performance when demand and cost parameters are time varying. The first model minimizes the total cost for opening and operating facilities and the associated transportation costs when demand and cost parameters are time varying. The model is solved using Benders? decomposition. We show that in the presence of high relocation costs of facilities (opening and closing costs), this model can be solved as a special case by the dynamic location model. The second model minimizes the maximum regret or opportunity loss between a robust configuration of facilities and the optimal configuration for each time period. We implement local search and simulated annealing metaheuristics to efficiently obtain near optimal solutions for this model.

Combinatorial problems are ubiquitous in our society and solving such problems efficiently is often crucial. One technique for solving combinatorial problems is constraint-based local search. Its compositional nature together with its efficiency on large problem instances have made this technique particularly attractive. In this thesis we contribute to simplifying the solving of combinatorial problems using constraint-based local search. To provide higher-level modelling options, we introduce set variables and set constraints in local search by extending relevant local search concepts. We also propose a general scheme to follow in order to define what we call natural and balanced constraint measures, and accordingly define such measures for over a dozen set constraints. However, defining such measures for a new constraint is time-consuming and error-prone. To relieve the user from this, we provide generic measures for any set constraint modelled in monadic existential second-order logic. We also theoretically relate these measures to our proposed general scheme, and discuss implementation issues such as incremental algorithms and their worst-case complexities. To enable higher-level search algorithms, we introduce constraint-directed neighbourhoods in local search by proposing new constraint primitives for representing such neighbourhoods. Based on a constraint, possibly modelled in monadic existential second-order logic, these primitives return neighbourhoods with moves that are known in advance to achieve a decrease (or preservation, or increase) of the constraint measures, without the need to iterate over any other moves. We also present a framework for constraint-based local search where one can model and solve combinatorial problems with set variables and set constraints, use any set constraint modelled in monadic existential second-order logic, as well as use constraint-directed neighbourhoods. Experimental results on three real-life problems show the usefulness in practice of our theoretical results: our running times are comparable to the current state-of-the-art approaches to solving the considered problems.

La traduction statistique vise l’automatisation de la traduction par le biais de modèles statistiques. Dans ce travail, nous relevons un des grands défis du domaine : la recherche (Brown et al., 1993). Les systèmes de traduction statistique de référence, tel Moses (Koehn et al., 2007), effectuent généralement la recherche en explorant l’espace des préfixes par programmation dynamique, une solution coûteuse sur le plan computationnel pour ce problème potentiellement NP-complet (Knight, 1999). Nous postulons qu’une approche par recherche locale (Langlais et al., 2007) peut mener à des solutions tout aussi intéressantes en un temps et un espace mémoire beaucoup moins importants (Russell et Norvig, 2010). De plus, ce type de recherche facilite l’incorporation de modèles globaux qui nécessitent des traductions complètes et permet d’effectuer des modifications sur ces dernières de manière non-continue, deux tâches ardues lors de l’exploration de l’espace des préfixes. Nos expériences nous révèlent que la recherche locale en traduction statistique est une approche viable, s’inscrivant dans l’état de l’art. / Statistical machine translation is a concerted effort towards the automation of the translation process. In the work presented here, we explore one of the major challenges of statistical machine translation: the search step (Brown et al., 1993). State of the art systems such as Moses (Koehn et al., 2007) search by exploring the prefix search space, a computationally costly solution to this potentially NP-complete problem (Knight, 1999). We propose that a local search approach can yield solutions which are qualitatively just as interesting, while keeping memory space and execution time at lower levels (Russell et Norvig, 2010). Furthermore, this type of search facilitates the use of global models for which a complete translation is needed and allows for non-continuous modifications, two tasks made difficult by exploring the prefix search space. The experiments we have conducted reveal that the use of local search during the search step in statistical machine translation is a viable, state of the art approach.

[pt] A exploração e produção (EeP) offshore de óleo e gás envolve várias operações complexas e importantes, como perfuração, avaliação, completação e manutenção de poços. A maioria dessas tarefas requer o uso de sondas, um recurso custoso e escasso que as companhias de petróleo precisam planejar e programar corretamente. Na literatura, este problema é chamado de Programação de Sondas. Todavia, existem poucos estudos relacionados aos poços marítimos e às atividades de perfuração e nenhum destes com funções objetivo e restrições realistas, como orçamento. Por isso, muitas empresas de petróleo têm fortes dificuldades no planejamento das sondas, resultando em grandes custos para elas. Com o objetivo de preencher essa lacuna, esta dissertação estuda um problema de programação de sondas em uma empresa petroleira e propõe um método híbrido para determinar a frota de sondas e seu cronograma, que minimize o orçamento da empresa. Dois modelos de programação matemática – um para minimização das sondas e outro para minimizar seu orçamento com variações da unidade de tempo utilizada (dia ou semana) – e várias heurísticas – usando algoritmos de busca local e variable neighborhood descent (VND) com três estruturas de vizinhança e duas estratégias de busca (first e best improvemment) e métodos construtivos- foram desenvolvidos e testados em duas instâncias (uma pequena e uma grande), baseadas em dados reais da empresa do caso de estudo. As três estruturas de vizinhanças são baseadas em movimentos de insert, uma delas não permite alterar as datas de alocação das tarefas na solução inicial, outra permite adiar tarefas e a última as posterga.Os resultados indicaram a dificuldade no desempenho dos modelos matemáticos nas grandes instâncias e uma forte capacidade das heurísticas para encontrar soluções similares com muito menos esforço computacional. Na instância pequena, o modelo exato para minimizar o orçamento encontrou soluções um pouco melhores que a heurística (diferença de entre 0,4 por cento e 5,6 por cento), embora necessitando de mais esforço computacional, principalmente os modelos com unidades de tempo em dias. Porém, na instância maior, as soluções da programação matemática possuíram altos gap (mais de 11 por cento) e altos tempos computacionais (pelo menos 12 horas), tendo o modelo matemático mais completo sido incapaz de encontrar soluções inteiras viáveis ou limites inferiores depois de mais de um dia rodando. Enquanto isso, as heurísticas foram capazes de encontrar soluções similares ou até melhores (desvios de -6 por cento e 14 por cento em relação a melhor solução exata) em um tempo muito menor, tendo 70 das 156 heurísticas desenvolvidas superado os modelos matemáticos. Além disso, os melhores resultados heurísticos foram utilizando algoritmos de variable neighborhood descent (VND) com estruturas de vizinhanças que realizavam movimentos de insert de tarefas em sondas existentes ou novas e permitiam postergar ou adiantar as tarefas das sondas. A abordagem hibrida foi comparada também com uma abordagem puramente heurística, tendo a primeira obtido melhores resultados. Por fim, os resultados demonstram que o método híbrido proposto combinando o modelo matemático que minimiza o número de sondas com as heurísticas de busca local é uma ferramenta de suporte a decisão rápida e prática, com potencial para reduzir milhões de dólares para as empresas petroleiras do mercado offshore, com capacidade para encontrar cronogramas próximos da solução ótima com pouco esforço computacional, mesmo em instâncias grandes onde a maioria dos métodos exatos é muito complexa e lenta. / [en] The offshore exploration and production (EandP) of Oil and Gas involves several complex and important operations and relies, mostly, in the use of rigs, a scarce and costly resource that oil companies need to properly plan and schedule. In the literature, this decision is called the Rig Scheduling Problem (RSP). However, there is not any study related to offshore wells and drilling activities with realistic objective functions. Aiming to fulfill this gap, this dissertation studies a rig scheduling problem of a real offshore company and proposes a matheuristic approach to determine a rigs fleet and schedule that minimizes the budget. Two mathematical models – one for rigs fleet minimization and another one that minimizes the rigs budget – and several heuristics – using local search (LS) and variable neighborhood descent (VND) algorithms with three neighborhood structures and also constructive methods – were developed and tested in two instances based on real data of the studied company. In the small instance, the programming model found slightly better solutions than the heuristic, despite requiring more computational effort. Nevertheless, in the large instance, the mathematical programming solutions present large gaps (over 11 percent) and an elevated computational time (at least 12 hours), while the heuristics can find similar (or even better) solutions in a shorter time (minutes), having 70 of 156 heuristics outperformed the mathematical models. Last, the matheuristic combination of the simplest mathematical model with the heuristics has found the best known solutions (BKS) of the large instance with a moderate computational effort.

[pt] A dissertação aborda um problema real de reprogramação de uma frota de embarcações do tipo PLSV (Pipe Laying Support Vessel), responsáveis pelas interligações de poços petrolíferos submarinos. O cronograma de curto prazo dessas embarcações está sujeito à inúmeras incertezas inerentes às operações realizadas, acarretando em ociosidade nas embarcações ou postergações na produção de petróleo, que podem resultar em prejuízo de milhões de reais. Uma metaheurística ILS (Iterated Local Search) é proposta para atender a frequente demanda por reprogramações dos PLSVs. O método é composto de uma fase inicial de viabilização, para tratar potenciais inconsistências nas programações. Na sequência, iterativamente, são realizadas perturbações na solução por meio de movimentos de swap e aplicada uma busca local baseada na vizinhança insert, a fim de fugir de ótimos locais e encontrar soluções que aprimorem o cronograma. Foram feitos experimentos com diferentes parâmetros e critérios do ILS, sendo definidas duas abordagens aplicadas a dez instâncias oriundas de uma programação real de PLSVs. A partir de uma função de avaliação, capaz de medir o impacto operacional na programação, o ILS proporcionou uma melhoria média nos cronogramas acima de 91 por cento, quando comparados aos cronogramas originais. As soluções foram obtidas em um tempo computacional médio de 30 minutos, aderente ao processo da companhia. Em função dos resultados alcançados, o método provou ser uma boa base para uma ferramenta de apoio à decisão para a reprogramação dos PLSVs. / [en] This dissertation addresses a real-life rescheduling problem of a Pipe Laying Support Vessels (PLSVs) fleet, in charge of subsea oil wells interconnections. The short-term schedule of these vessels is subject to uncertainties inherent to its operations, resulting in ships idleness or delays in oil production, which may lead to losses of millions of Brazilian Reais. A method based on the ILS (Iterated Local Search) metaheuristic is proposed to meet the frequent demand of PLSVs rescheduling. The first step of this method aims to find a feasible initial solution from an incoming schedule with potencial inconsistencies. The following steps consists in, iteratively, performing a perturbation on a solution through swap movements and applying a local search based on the insertion neighborhood, in order to escape from local optimal and find better solutions. Extensive preliminary experiments were conducted considering different ILS parameters setups. The two most performing setups were selected and applied to ten instances of a real PLSV schedule. Taking into account an objective function that measures the operational impact on schedules, the ILS provided an average improvement above 91 percent in schedules when compared to the original planning. These solutions were obtained in an average computational time of 30 minutes, which fits in the company process. The obtained results showed that the proposed method might be a basis for a decision support tool for the PLSVs rescheduling problem.

This thesis presents two new routing problems, both with objective functions focused on relative positioning of teams during the routing horizon. The relative positioning results in temporal and spatial dependencies among routes and is quantified with a nonlinear dispersion metric, designed to evaluate the instantaneous distances among teams over a time interval. This metric allows the design of objective functions to approximate teams during routes execution, when minimized, or disperse them, when maximized. Both approximation and dispersion are important routing characteristics in some practical applications, and two new optimization problems are proposed with these opposite objectives. The first one is a variation of the Multiple Traveling Salesman Problem, and its goal is to find a set of tours where the salesmen travel close to each other, minimizing dispersion. A Local Search Genetic Algorithm is proposed to solve the problem. It includes specialized genetic operators and neighborhoods. A new set of benchmark instances is proposed, adapted for the new problem from literature instances. Computational results show that the proposed approach provides solutions with the desired characteristics of minimal dispersion. The second problem is a bi-objective arc routing problem in which routes must be constructed in order to maximize collected profit and dispersion of teams. The maximization of the dispersion metric fosters the scattering of the teams during routing procedure. Usually, profit and dispersion objectives are conflicting, and by using a bi-objective approach the decision maker is able to choose a trade-off between collecting profits and scattering teams. Two solution methods are proposed, a Multi-objective Genetic Algorithm and a Multi-objective Genetic Local Search Algorithm, both specialized in order to exploit the characteristics of the problem. It is demonstrated, by means of computational experiments on a new set of benchmark instances, that the proposed approach provides approximation sets with the desired characteristics. / Esta tese apresenta dois novos problemas de roteamento, ambos com funções objetivo voltadas para o posicionamento relativo das equipes durante o horizonte de roteamento. O posicionamento relativo resulta em uma dependência temporal e espacial entre rotas e é quantificado com uma métrica de dispersão não-linear, projetada para avaliar as distâncias instantâneas entre as equipes ao longo de um intervalo de tempo. Esta métrica permite a concepção de funções objetivo para aproximar as equipes durante a execução das rotas, quando minimizada, ou para dispersá-las, quando maximizada. Tanto a aproximação quanto a dispersão são características importantes de roteamento em algumas aplicações práticas, e dois novos problemas de otimização são propostos com esses objetivos opostos. O primeiro é uma variação do Problema de Múltiplos Caixeiros Viajantes, e seu objetivo é encontrar um conjunto de rotas em que os caixeiros viajam próximos uns dos outros, minimizando a dispersão. Um Algoritmo Genético com Busca Local é proposto para resolver o problema. Ele inclui operadores genéticos e vizinhanças especializados. Um novo conjunto de instâncias é proposto, adaptado para o novo problema de instâncias da literatura. Resultados computacionais mostram que a abordagem proposta proporciona soluções com as características desejadas de dispersão mínima. O segundo problema é um problema de roteamento de arcos biobjetivo em que as rotas devem ser construídas de modo a maximizar o lucro recolhido e o distanciamento entre as equipes. A maximização da métrica promove a dispersão das equipes durante a execução das rotas. Normalmente, os objetivos de lucro e dispersão são conflitantes, e com uma abordagem biobjetivo o tomador de decisão é capaz de avaliar a troca entre a coleta de lucros e a dispersão de equipes. Dois métodos de solução são propostos, um Algoritmo Genético Multiobjetivo e um Algoritmo Genético Multiobjetivo com Busca Local, ambos especializados para explorar as características do problema. É demonstrado, por meio de experimentos computacionais sobre um novo conjunto de instâncias, que a abordagem proposta fornece conjuntos de aproximação com as características desejadas.

The following thesis deals with optimization of freight transport planning. The goal is to minimize expenses connected to transportation, which emerge from travelled distance. The expenses can be heavily reduced, if the routes are correctly planned, especially when there is a large number of customers to be served. This thesis focuses on solving the problem by using the evolutional algorithms, that are optimization methods based on principles of evolution. Thesis concentrates on Heterogeneous Fixed Fleet Vehicle Routing Problem. Thesis introduces multiple evolutional algorithms and their results are compared. The best algorithm, evolutional strategy with local neighbourhood search, achieves similar, for certain tasks even better results, than other existing evolutional algorithms, created to solve given problem.

Thesis deals with discrete optimization problems. It focusses on faster ways to find good solutions by means of heuristics and parallel processing. Based on ant colony optimization (ACO) algorithm coupled with k-optimization local search approach, it aims at massively parallel computing on graphics processors provided by Nvidia CUDA platform. Well-known travelling salesman problem (TSP) is used as a case study. Solution is based on dividing task into subproblems using tour-based partitioning, parallel processing of distinct parts and their consecutive recombination. Provided parallel code can perform computation more than seventeen times faster than the sequential version.

De nombreux problèmes d'optimisation combinatoire considèrent plusieurs objectifs, souvent conflictuels. Cette thèse s'intéresse à l'utilisation de méthodes de recherche locale, de structures de données et de recherche Monte-Carlo pour la recherche de l'ensemble des solutions efficaces de tels problèmes, représentant l'ensemble des meilleurs compromis pouvant être réalisés en considération de tous les objectifs.Nous proposons une nouvelle méthode d'approximation appelée 2-Phase Iterated Pareto Local Search based on Decomposition (2PIPLS/D) combinant les concepts de recherche locale Pareto (PLS) et de décomposition. La PLS est une descente de recherche locale adaptée au multi-objectif, et la décomposition consiste en la subdivision du problème multi-objectif en plusieurs problèmes mono-objectif. Deux méthodes d'optimisation mono-objectif sont considérées: la recherche locale itérée et la recherche Monte-Carlo imbriquée. Deux modules principaux sont intégrés à 2PIPLS/D. Le premier généralise et améliore une méthode existante et génère un ensemble initial de solutions. Le second réduit efficacement l'espace de recherche et permet d'accélérer la PLS sans négliger la qualité de l'approximation générée. Nous introduisons aussi deux nouvelles structures de données gérant dynamiquement un ensemble de solutions incomparables, la première est spécialisée pour le cas bi-objectif et la seconde pour le cas général.2PIPLS/D est appliquée au Problème du Voyageur de Commerce bi-objectif et tri-objectif et surpasse ses concurrents sur les instances testées. Ensuite, 2PIPLS/D est appliquée à un nouveau problème avec cinq objectifs en lien avec la récente réforme territoriale d'agrandissement des régions françaises. / Many Combinatorial Optimization problems consider several, often conflicting, objectives. This thesis deals with Local Search, data structures and Monte Carlo Search methods for finding the set of efficient solutions of such problems, which is the set of all best possible trade-offs given all the objectives.We propose a new approximation method called 2-Phase Iterated Pareto Local Search based on Decomposition (2PIPLS/D) combining the notions of Pareto Local Search (PLS) and Decomposition. PLS is a local search descent adapted to Multi-Objective spaces, and Decomposition consists in the subdivision of the Multi-Objective problem into a number of Single-Objective problems. Two Single-Objective methods are considered: Iterated Local Search and Nested Monte Carlo Search. Two main components are embedded within the 2PIPLS/D framework. The first one generalizes and improves an existing method generating an initial set of solutions. The second one reduces efficiently the search space and accelerates PLS without notable impact on the quality of the generated approximation. We also introduce two new data structures for dynamically managing a set of incomparable solutions. The first one is specialized for the bi-objective case, while the second one is general.2PIPLS/D is applied to the bi-objective and tri-objective Traveling Salesman Problem and outperforms its competitors on tested instances. Then, 2PIPLS/D is instantiated on a new five-objective problem related to the recent territorial reform of French regions which resulted in the reassignment of departments to new larger regions.