Spanning tree optimization problems are related to many practical applications. Several of these problems are NP-Hard, which limits the utility of exact methods and can require alternative approaches, like metaheuristics. A common issue for many metaheuristics is the data structure used to represent and manipulate the solutions. A data structure with efficient operations can expand the usefulness of a method by allowing larger instances to be solved in a reasonable amount of time. We propose the 2LETT data structure and uses it to represent spanning trees in two metaheuristics: mutation-based evolutionary algorithms and local search algorithms. The main operation of 2LETT is the exchange of one edge in the represented tree by another one, and it has O(√n) time, where n is the number of vertices in the tree. We conducent qualitative and quantitative evaluations for 2LETT and other structures in the literature. For the main operation of edge exchange in evolutionary algorithms, the computational experiments show that 2LETT has the best performance for trees with more than 10,000 vertices. For local search algorithms, 2LETT is the best option to deal with large trees with large diameters. / Os problemas de otimização de árvores geradoras estão relacionados a muitas aplicações práticas. Vários desses problemas são NP-difícies, o que limita a utilidade de métodos exatos e pode exigir abordagens alternativas, como metaheurísticas. Um questão relevante para muitas metaheurísticas é a estrutura de dados usada para representar e manipular as soluções. Uma estrutura de dados com operações eficientes pode aumentar a utilidade de um método, permitindo que instâncias maiores sejam resolvidas em um período de tempo razoável. Propomos a estrutura de dados 2LETT e a usamos para representar árvores geradoras em duas metaheurísticas: algoritmos evolutivos baseados em mutações e algoritmos de busca local. A operação principal da 2LETT é a troca de uma aresta na árvore representada por outra aresta. Esta operação tem tempo de O(√n), onde n é o número de vértices na árvore. Conduzimos avaliações qualitativas e quantitativas para 2LETT e outras estruturas na literatura. Para a principal operação de troca de arestas em algoritmos evolutivos, os experimentos computacionais mostram que a 2LETT possui o melhor desempenho para árvores com mais de 10.000 vértices. Para algoritmos de busca local, o 2LETT é a melhor opção para lidar com árvores grandes com grandes diâmetros.

L’objectif de cette thèse est le développement de méthodes d’optimisation multi-objectif pour la résolution de problèmes de conception des structures mécaniques. En effet, la plupart des problèmes réels dans le domaine de la mécanique des structures ont plusieurs objectifs qui sont souvent antagonistes. Il s’agit, par exemple, de concevoir des structures en optimisant leurs poids, leurs tailles, et leurs coûts de production. Le but des méthodes d’optimisation multi-objectif est la recherche des solutions de compromis entre les objectifs étant donné l’impossibilité de satisfaire tout simultanément. Les métaheuristiques sont des méthodes d’optimisation capables de résoudre les problèmes d’optimisation multi-objective en un temps de calcul raisonnable sans garantie de l’optimalité de (s) solution (s). Au cours des dernières années, ces algorithmes ont été appliqués avec succès pour résoudre le problème des mécaniques des structures. Dans cette thèse deux métaheuristiques ont été développées pour la résolution des problèmes d’optimisation multi-objectif en général et de conception de structures mécaniques en particulier. Le premier algorithme baptisé MOBSA utilise les opérateurs de croisement et de mutation de l’algorithme BSA. Le deuxième algorithme nommé NNIA+X est une hybridation d’un algorithme immunitaire et de trois croisements inspirés de l’opérateur de croisement original de l’algorithme BSA. Pour évaluer l’efficacité et l’efficience de ces deux algorithmes, des tests sur quelques problèmes dans littérature ont été réalisés avec une comparaison avec des algorithmes bien connus dans le domaine de l’optimisation multi-objectif. Les résultats de comparaison en utilisant des métriques très utilisées dans la littérature ont démontré que ces deux algorithmes peuvent concurrencer leurs prédécesseurs. / The objective of this thesis is the development of multi-objective optimization methods for solving mechanical design problems. Indeed, most of the real problems in the field of mechanical structures have several objectives that are often antagonistic. For example, it is about designing structures by optimizing their weight, their size, and their production costs. The goal of multi-objective optimization methods is the search for compromise solutions between objectives given the impossibility to satisfy all simultaneously. Metaheuristics are optimization methods capable of solving multi-objective optimization problems in a reasonable calculation time without guaranteeing the optimality of the solution (s). In recent years, these algorithms have been successfully applied to solve the problem of structural mechanics. In this thesis, two metaheuristics have been developed for the resolution of multi-objective optimization problems in general and of mechanical structures design in particular. The first algorithm called MOBSA used the crossover and mutation operators of the BSA algorithm. The second one named NNIA+X is a hybridization of an immune algorithm and three crossover inspired by the original crossover operator of the BSA algorithm. To evaluate the effectiveness and efficiency of these two algorithms, tests on some problems in literature have been made with a comparison with algorithms well known in the field of multi-objective optimization. The comparison results using metrics widely used in the literature have shown that our two algorithms can compete with their predecessors.

Une tâche importante qui a été étudiée dans le contexte de données à forte dimensionnalité est la tâche connue sous le nom de subspace clustering. Le subspace clustering est généralement reconnu comme étant plus compliqué que le clustering standard, étant donné que cette tâche vise à détecter des groupes d’objets similaires entre eux (clusters), et qu’en même temps elle vise à trouver les sous-espaces où apparaissent ces similitudes. Le subspace clustering, ainsi que le clustering traditionnel ont été récemment étendus au traitement de flux de données en mettant à jour les modèles de clustering de façon incrémentale. Les différents algorithmes qui ont été proposés dans la littérature, reposent sur des bases algorithmiques très différentes. Parmi ces approches, les algorithmes évolutifs ont été sous-explorés, même si ces techniques se sont avérées très utiles pour traiter d’autres problèmes NP-difficiles. L’objectif de cette thèse a été de tirer parti des nouvelles connaissances issues de l’évolution afin de concevoir des algorithmes évolutifs qui traitent le problème du subspace clustering sur des jeux de données statiques ainsi que sur des flux de données dynamiques. Chameleoclust, le premier algorithme développé au cours de ce projet, tire partie du grand degré de liberté fourni par des éléments bio-inspirés tels qu’un génome de longueur variable, l’existence d’éléments fonctionnels et non fonctionnels et des opérateurs de mutation incluant des réarrangements chromosomiques. KymeroClust, le deuxième algorithme conçu dans cette thèse, est un algorithme de k-medianes qui repose sur un mécanisme évolutif important: la duplication et la divergence des gènes. SubMorphoStream, le dernier algorithme développé ici, aborde le problème du subspace clustering sur des flux de données dynamiques. Cet algorithme repose sur deux mécanismes qui jouent un rôle clef dans l’adaptation rapide des bactéries à des environnements changeants: l’amplification de gènes et l’absorption de matériel génétique externe. Ces algorithmes ont été comparés aux principales techniques de l’état de l’art, et ont obtenu des résultats compétitifs. En outre, deux applications appelées EvoWave et EvoMove ont été développés pour évaluer la capacité de ces algorithmes à résoudre des problèmes réels. EvoWave est une application d’analyse de signaux Wi-Fi pour détecter des contextes différents. EvoMove est un compagnon musical artificiel qui produit des sons basés sur le clustering des mouvements d’un danseur, décrits par des données provenant de capteurs de déplacements. / An important task that has been investigated in the context of high dimensional data is subspace clustering. This data mining task is recognized as more general and complicated than standard clustering, since it aims to detect groups of similar objects called clusters, and at the same time to find the subspaces where these similarities appear. Furthermore, subspace clustering approaches as well as traditional clustering ones have recently been extended to deal with data streams by updating clustering models in an incremental way. The different algorithms that have been proposed in the literature, rely on very different algorithmic foundations. Among these approaches, evolutionary algorithms have been under-explored, even if these techniques have proven to be valuable addressing other NP-hard problems. The aim of this thesis was to take advantage of new knowledge from evolutionary biology in order to conceive evolutionary subspace clustering algorithms for static datasets and dynamic data streams. Chameleoclust, the first algorithm developed in this work, takes advantage of the large degree of freedom provided by bio-like features such as a variable genome length, the existence of functional and non-functional elements and mutation operators including chromosomal rearrangements. KymeroClust, our second algorithm, is a k-medians based approach that relies on the duplication and the divergence of genes, a cornerstone evolutionary mechanism. SubMorphoStream, the last one, tackles the subspace clustering task over dynamic data streams. It relies on two important mechanisms that favor fast adaptation of bacteria to changing environments, namely gene amplification and foreign genetic material uptake. All these algorithms were compared to the main state-of-the-art techniques, obtaining competitive results. Results suggest that these algorithms are useful complementary tools in the analyst toolbox. In addition, two applications called EvoWave and EvoMove have been developed to assess the capacity of these algorithms to address real world problems. EvoWave is an application that handles the analysis of Wi-Fi signals to detect different contexts. EvoMove, the second one, is a musical companion that produces sounds based on the clustering of dancer moves captured using motion sensors.

O restabelecimento de energia em sistemas de distribuição de energia elétrica radiais geralmente envolve a reconfiguração de redes para restaurar eletricidade à(s) área(s) fora de serviço. As principais técnicas para restabelecimento de energia em sistemas de distribuição de grande porte têm sido os algoritmos evolutivos (AEs). Após a falta ter sido identificada e a zona em falta ter sido isolada do sistema, o algoritmo deve encontrar soluções em que: 1) supra com energia o maior número de consumidores possível, 2) minimize o número de operações de chaveamentos, 3) não viole restrições operacionais do sistema, 4) reduza o total de perdas resistivas, 5) a configuração da rede seja radial e, 6) obtenha tal solução em tempo real. Este projeto emprega uma nova estrutura de dados para manipular grafos produzindo exclusivamente configurações radiais e conexas, chamada representação nó-profundidade (RNP), garantindo que todas as soluções potenciais geradas pelo algoritmo satisfaçam os itens (1) e (5). Além disso, propõe-se um AE utilizando a RNP capaz de encontrar planos de restabelecimento adequados para sistemas de distribuição de larga-escala, com milhares de chaves e barras, em tempo real. / Energy restoration in radial distribution systems usually involves the network reconfiguration to restore the electricity to the out-of-service areas. The main approaches for energy restoration in large-scale distribution systems have been the evolutionary algorithms (EAs). After a fault has been identified and isolated, the algorithm must find solutions that: 1) supply energy to the larger number of consumers, 2) reduce the number of switching operations, 3) respect operational constraints of the system, 4) reduce the amount of power losses, 5) generate exclusively radial configurations and 6) find solutions in real time. This work uses a new data structure, called node-depth encoding (NDE), to manipulate graphs producing exclusively radial and connected configurations, and guaranteeing that all potential solutions generated by the algorithm satisfy items (1) and (5). Moreover, we propose an EA using the NDE that is capable of finding adequate restoration plans in real time for large-scale distribution systems, with thousands of switches and buses.

A tendência crescente na geração de harmônicos nos sistemas elétricos de potência tem ganhado atenção especial no planejamento das redes de transporte de energia elétrica, uma vez que os crescimentos observados acontecem em regiões que antes não representavam qualquer tipo de preocupação. Um dos principais fatores que contribuíram para esse novo contexto é a característica distribuída da geração de harmônicos. Devido a essa nova realidade, métodos mais aprimorados para avaliação de desempenho e modelos mais precisos para a representação de equipamentos tornaram-se necessários. Sendo assim, a pesquisa realizada para a elaboração da presente tese fundamentou a sua investigação em três tópicos com o intuído de fornecer contribuições que permitissem uma avaliação mais precisa das redes elétricas, proporcionando, assim, resultados mais aderentes com a realidade existente: Modelagem Agregada de Carga; Equivalentes de Redes; Estimação de Estados das Distorções Harmônicas. Através das contribuições feitas nesses tópicos, torna-se possível a consideração de aspectos que antes eram ignorados na avaliação harmônica das redes de transporte de energia elétrica e, assim, permite-se uma verificação precisa dos impactos da característica distribuída da geração de harmônicos nos sistemas elétricos de potência. / The growing rate of harmonic generation present in the electric power systems has gained special attention in the planning process of power networks. The major factor that contributed for this new context is the increasing harmonic generation observed in regions that did not use to represent any concern in the past. One of the main causes for this new trend is the distributed characteristic of the harmonic generation. In this new environment, sophisticated methods and models have become necessary, in order to precisely represent the electric elements behaviour and to accurately evaluate the systems performance. As a result, the research work presented in this thesis focused in three different topics, in order to provide contributions that would lead to a more accurate performance evaluation of the power networks and that would provide results closer to the values found in the field: Aggregate Load Modeling; Network Equivalents; Harmonic State Estimation. These contributions would allow the consideration of aspects that normally are ignored in the harmonic assessment of power systems. Consequently, the evaluation of the impacts caused by the distributed generation of harmonics becomes more accurate.

Métodos de identificação de pessoas sempre foram muito importantes para toda a sociedade. Atualmente, as pesquisas em biometria vêm sendo amplamente incentivadas por diversos setores da indústria mundial com o objetivo de melhorar ou substituir os atuais sistemas de segurança e de identificação de pessoas. O campo da biometria abarca uma grande variedade de tecnologias usadas para identificar e verificar a identidade de uma pessoa por meio da mensuração e análise de diversas características físicas e/ou comportamentais do ser humano. Diversas modalidades biométricas têm sido propostas para reconhecimento de pessoas, como impressão digital, íris, face e fala. Estas modalidades biométricas possuem características distintas em termos de desempenho, mensurabilidade e aceitabilidade. Uma questão a ser considerada com a aplicação biométrica é sua robustez a ataques por circunvenção, repetição e ofuscação. Esses ataques estão se tornando cada vez mais frequentes e questionamentos estão sendo levantados a respeito dos níveis de segurança das formas de reconhecimento. Sinais biomédicos como eletrocardiograma (ECG), eletroencefalograma (EEG) e eletromiograma (EMG) têm sido cada vez mais estudados e aplicados ao reconhecimento biométrico. Em específico, os sinais de ECG têm sido largamente adotados para o reconhecimento biométrico em diversos trabalhos. Por outro lado, análise de séries temporais tem sido usada com sucesso em muitas diferentes aplicações para identificar padrões temporais nos dados. Embora dinâmica simples possa ser observada com ferramentas analíticas tradicionais tais como transformada de fourier, transformada wavelet, a representação simbólica pode melhorar a análise de processos que são complexos e possivelmente caótico. Além disso, representação simbólica pode também reduzir a sensibilidade a ruído e melhorar bastante a eficiência computacional. No entanto, existem aspectos estruturais e paramétricos de projeto que podem conduzir a uma degradação de desempenho. Na ausência de uma metodologia sistemática e de baixo custo para a proposição de técnicas de representação simbólicas otimamente especificadas, os comitês de máquinas, mais especificamente ensemble, se apresentam como alternativas promissoras. Neste estudo, os componentes do ensemble, que correspondem as técnicas de representação simbólicas, e seus respectivos parâmetros foram selecionados via algoritmos evolutivos. O objetivo é explorar conjuntamente potencialidades advindas das técnicas de representação simbólicas e comitê de máquinas para reconhecimento biométrico baseado em sinais de ECG. Resultados experimentais conduzidos sobre dois conjuntos de dados disponíveis publicamente indicam que a abordagem proposta pode melhorar o desempenho do reconhecimento quando comparada com as técnicas tradicionais / Identification people methods have been very important for the whole society. Currently, research on biometrics have been widely encouraged by various sectors of the industry worldwide in order to improve or replace existing security systems and people identification. The field of biometrics includes a variety of technologies used to identify or verify the identity of a person by measuring and analyzing various physical and/or behavioral aspects of the human being. Several biometric methods have been proposed for recognition of people, such as fingerprint, iris, face and speech. These biometric modalities have different characteristics in terms of performance, measurability and acceptability. One issue to be considered with the biometric application in the real world is its robustness to attacks by circumvention, repetition and obfuscation. These attacks are becoming more frequent and more questions are being raised about the levels of security that this technology can offer. Biomedical signals such as electrocardiogram (ECG), electroencephalogram (EEG) and electromyogram (EMG) have been increasingly studied and applied to biometric recognition. Specifically, ECG signals have been widely adopted for biometric recognition in various works. On the other hand, time series analysis has been used successfully in many different applications to identify temporal patterns in the data. Although simple dynamics can be observed with traditional analytical tools such as fourier transform, wavelet transform, the symbolic representation can improve the analysis of processes that are complex and possibly chaotic. In addition, symbolic representation can also reduce noise sensitivity and greatly improve computational efficiency. However, there are structural and parametric design aspects that can lead to performance degradation. In the absence of a systematic and inexpensive methodology for proposing optimally specified symbolic representation techniques, machine committees, more specifically ensemble, present themselves as promising alternatives. In this study, the components of the committee, which correspond to techniques of symbolic representation, and their respective parameters were selected via evolutionary algorithms. The objective is to jointly explore the potentialities of both symbolic representation techniques and machine committee for biometric recognition based on ECG signals. Experimental results conducted on two publicly available datasets indicate that the proposed approach may improve recognition performance when compared to traditional techniques

A detecção de furtos e fraudes nos sistemas de energia provocados por consumidores irregulares é o principal alvo em análises de perdas não-técnicas ou comerciais pelas empresas de energia. Embora a identificação automática de perdas nãotécnicas tenha sido amplamente estudada, a tarefa de selecionar as características mais representativas em um grande conjunto de dados a fim de aumentar a taxa de acerto da identificação, bem como para caracterizar possíveis consumidores irregulares como um problema de otimização, não tem sido muito explorada neste contexto. Neste trabalho, visa-se o desenvolvimento de algoritmos híbridos baseados em técnicas evolutivas a fim de realizar a seleção de características no âmbito da caracterização de perdas não-técnicas, comparando as suas taxas de acerto e verificando as características selecionadas. Vários classificadores são comparados, com destaque para a técnica Floresta de Caminhos Ótimos por sua robustez, sendo ela a técnica escolhida para o cálculo da função objetivo das técnicas evolutivas, analisando o desempenho das mesmas. Os resultados demonstraram que a seleção de características mais representativas podem melhorar a taxa de acerto da classificação de possíveis perdas não-técnicas quando comparada à classificação sem o processo de seleção de características em conjuntos de dados compostos por perfis de consumidores industriais e comerciais. Isto significa que existem características que não são pertinentes e podem diminuir a taxa de acerto durante a classificação dos consumidores. Através da metodologia proposta com o processo de seleção de características, é possível caracterizar e identificar os perfis de consumidores com mais precisão, afim de minimizar os custos com tais perdas, contribuindo para a recuperação de receita das companhias de energia elétrica. / The detection of thefts and frauds in power systems caused by irregular consumers is the most actively pursued analysis in non-technical losses by electric power companies. Although non-technical losses automatic identification has been massively studied, the task of selecting the most representative features in a large dataset, in order to boost the identification accuracy, as well as characterizing possible irregular consumers as a problem of optimization, has not been widely explored in this context. This work aims at developing hybrid algorithms based on evolutionary algorithms in order to perform feature selection in the context of non-technical losses characterization. Although several classifiers have been compared, we have highlighted the Optimum-Path Forest (OPF) technique mainly because of its robustness. Thus, the OPF classifier was chosen to compute the objective function of evolutionary techniques, analyzing their performances. This procedure with feature selection is compared with the procedure without feature selection in datasets composed by industrial and commercial consumers profiles. The results demonstrated that selecting the most representative features can improve the classification accuracy of possible non-technical losses. This means that there are irrelevant features and they can reduce the classification accuracy of consumers. Considering the methodology proposed with feature selection procedure, it is possible to characterize and identify consumer profiles more accurately, in order to minimize costs with such losses, contributing to the recovery of revenue from electric power companies.

Este trabalho foca o desenvolvimento de algoritmos de otimização para o problema de PSP puramente ab initio. Algoritmos que melhor exploram o espaço de potencial de soluções podem, em geral, encontrar melhores soluções. Esses algoritmos podem beneficiar ambas abordagens de PSP, tanto o modelo ab initio quanto os baseados em conhecimento a priori. Pesquisadores tem mostrado que Algoritmos Evolutivos Multiobjetivo podem contribuir significativamente no contexto do problema de PSP puramente ab initio. Neste contexto, esta pesquisa investiga o Algoritmo Evolutivo Multiobjetivo baseado em Tabelas aplicado ao PSP puramente ab initio, que apresenta interessantes resultados para proteínas relativamente simples. Por exemplo, um desafio para o PSP puramente ab initio é a predição de estruturas com folhas-. Para trabalhar com tais proteínas, foi desenvolvido procedimentos computacionalmente eficientes para estimar energias de ligação de hidrogênio e solvatação. Em geral, estas não são consideradas no PSP por abordagens que combinam métodos de otimização e conhecimento a priori. Considerando somente van der Waals e eletrostática, as duas energias de interação que mais contribuem para a definição da estrutura de uma proteína, com as energias de ligação de hidrogênio e solvatação, o problema de PSP tem quatro objetivos. Problemas combinatórios (tais como o PSP), com mais de três objetivos, geralmente requerem métodos específicos capazes de lidar com muitos critérios. Para resolver essa limitação, este trabalho propõe um novo método para a otimização dos muitos objetivos, chamado Algoritmo Evolutivo Multiobjetivo com Muitas Tabelas (AEMMT). Esse método executa uma amostragem mais adequada do espaço de funções objetivo e, portanto, pode mapear melhor as regiões promissoras deste espaço. A capacidade de lidar com muitos objetivos capacita o AEMMT a utilizar melhor a informação oriunda das energias de solvatação e de ligação de hidrogênio, e então predizer estruturas com folhas- e algumas proteínas relativamente mais complexas. Do ponto de vista computacional, o AEMMT é um novo método que lida com muitos objetivos (mais de dez) encontrando soluções relevantes / This work focuses on the development of optimization algorithms for the purely ab initio Protein Structure Prediction (PSP) problem. Algorithms that better explore the space of potential solutions can in general find better solutions. Such algorithms can benefit both ab initio and template-based PSP, that uses priori knowledge. Researches have shown that Multiobjective evolutionary algorithms can contribute significantly in the context of purely ab initio PSP. In this context, this research investigates the Multiobjective Evolutionary Algorithm based on Tables applied to purely ab initio PSP, which has shown interesting results for relatively simple proteins. For example, one challenge for purely ab initio PSP is the prediction of structures with -sheets. To work with such proteins, this research has developed computationally efficient procedures to estimate hydrogen bond and solvation energies. In general, they are not considered by PSP approaches combining optimization methods with priori knowledge. Only by considering van der Waals and electrostatic, the two interaction energies that mostly contribute to defining a protein structure, and the hydrogen bond and solvation energies, the PSP problem has four objectives. Combinatorial problems (such as the PSP) with more than three objective usually require specific methods capable of dealing with many goals. To address this limitation, we propose a new method for many objective optimization, called Multiobjective Evolutionary Algorithm with Many Tables (MEAMT). This method performs a more adequate sampling of the space of objective functions and, therefore, can better map the promising regions of this space. The ability of dealing with many objectives enables the MEANT to better use information generated by solvation and hydrogen bond energies, and then predict structures with -sheets and some relatively complex proteins. From the computational point of view, the MEAMT is a new method for dealing with many objectives (more than ten) finding relevant solutions

Für die erfolgreiche Lösung eines Optimierungsproblems ist die Wahl der verwendeten Suchstrategie von entscheidender Bedeutung. Die vorliegende Arbeit untersucht die Kriterien dieser Wahl. Dabei stellen sich drei grundlegende Fragen: Welche Strategien der Optimierung eines gegebenen Problems existieren überhaupt, und was für Eigenschaften besitzen sie? Wodurch wird der Charakter eines Optimierungsproblems bestimmt, und gibt es Klassen ähnlicher Probleme? Besteht eine Verbindung zwischen den Eigenschaften der Strategien und den Klassen der Probleme, die es ermöglicht, für jede Problemklasse eine geeignete Optimierungsstrategie anzugeben? Dazu wird zuerst die Klasse der Evolutionären Algorithmen naher betrachtet, deren generelles Verhalten die Boltzmannstrategie, Darwinstrategie oder Boltzmann-Darwin-Strategie beschreiben. Als weiteres Beispiel wird die Multitome Strategie untersucht. In ihr wird das Problem unter verschiedenen Gesichtspunkten betrachtet und in Einzelanforderungen zerlegt, die abwechselnd optimiert werden. Für den speziellen Fall der Dichotomen Strategie wird die allgemeine zeitabhängige Lösung mit Hilfe der Methode der Charakteristiken bestimmt. Zur Beantwortung der zweiten Frage wird die Zustandsdichte als klassifizierende Größe des Optimierungsproblems eingeführt. Sie kann unter Verwendung der Boltzmannstrategie während des Optimierungslaufes durch zwei allgemeine Approximationsmethoden: die Methode der stationären Verteilungen und die Eigenvektormethode bestimmt werden. Aus der Zustandsdichte erhält man den Wirkungsgrad der Zufallssuche. Er charakterisiert den Ordnungsgrad des Problems und stellt damit ein wichtiges Maß der Problemschwierigkeit dar. Die entscheidende dritte Frage wird für Probleme der Optimierung frustrierter Sequenzen, der Netzwerkoptimierung und für das Faltungsproblem der RNA behandelt. Mit der Einführung der Klassen gerichteter und ungerichteter Strategien, die für Optimierungsprobleme mit niedrigem bzw. hohem Wirkungsgrad der Zufallssuche effektiv sind, kann eine Verbindung zwischen dem Strategieverhalten und dem Problemcharakter hergestellt werden, die es ermöglicht, für eine konkrete Optimierungsaufgabe die Klasse der geeigneten Strategien zu wählen. / A crucial point of successful solving an optimization problem is the choice of the used strategy. The present paper investigates the criteria of this choice. Thereby three fundamental questions put themselves: Which strategies of the optimization of a given problem exist altogether, and which properties characterize the strategies? How is the character of an optimization problem determined, and are there classes of similar problems? Does a combination exist between the characteristics of the strategies and the classes of problems, which makes it possible to indicate a suitable strategy for each class? The class of the Evolutionary Algorithms is considered more closely. The general behavior of the algorithms can be described by the Boltzmann strategy, Darwin strategy or Boltzmann-Darwin strategy. As a further example the Multitomic strategy is explored. In this approach the problem is considered under different points of view and decomposes in single demands, which are optimized alternately. For the special case of the Dichotomic strategy the general time dependent solution is determined. To answer the second question the density of states is introduced as classifying measure of optimization problems. The density can be determined during the optimization course by two general approaches: the method of the stationary distribution and the eigenvalue method. From the density of states one receives the efficiency of the random search. It describes the degree of order of the problem and presents an measure of the problem difficulty. The important third question is treated for problems of the optimization of frustrated sequences, the network optimization and RNA folding. The introduction of the classes of directed and non directed strategies, which are effective for problems with low and high efficiency of the random search, establishes a connection between the strategy and the character of the problem, which makes it possible to choose the class of the suitable strategies for a given optimization task.

L’objectif de ce projet de trois ans est de proposer des avancées conceptuelles et technologiques dans la résolution de problèmes d’ordonnancement du personnel. L’atteinte de cet objectif passe par la proposition de nouveaux algorithmes basés sur les métaheuristiques et leur implémentation sur les architectures de calcul haute performance. Ce projet s’inscrit en complémentarité du projet HORUS qui bénéficie d’une subvention ANR et qui réunit les expertises scientifiques de deux laboratoires universitaires spécialisés en optimisation et en calcul parallèle : l’équipe SysCom du laboratoire CReSTIC de l’URCA et l’équipe CaRO du laboratoire PRiSM de l’UVSQ. Les avancées technologiques proposées s’appuient également sur les moyens de calcul haute performance offerts par le Centre de Calcul Régional Champagne-Ardenne. / .Many academic and industrial optimization problems are multi-objective and have been of particular interest to researchers in recent years. These problems usually do not have a single optimal solution but a set of best trade-off solutions which form the so-called Pareto front in the objective space. In order to approximate the Pareto front, multi-objective evolutionary algorithms (MOEAs) have been largely investigated in the fields of continuous and combinatorial optimization. Contrary to some classical algorithms, MOEAs have the ability to provide a number of solutions in one single run and are less sensitive to the shape of the Pareto front.As they often require a high amount of computing resources to explore large portions of the search space and handle complex real-life constraints, MOEAs could greatly benefit from today's high-performance computing architectures. Although significant progress has been made in recent years in the design and improvement of parallel models for evolutionary algorithms, most of these models have limited scalability and ability to solve various problems. In fact, solving multi-objective combinatorial optimization problems efficiently on a large number of processors remains a challenge today.This thesis aims to propose an island model which is based on objective space division. The main features of the proposed model are the following (i) An organizer has a global view of the current search via a global archive (ii) Asynchronous cooperation between islands, especially for the exchange of local archives with the organizer to limit model overheads (iii)Control islands to guide the exploration of the search space and improve diversity (iv) A periodic use of a specific local search procedure to improve convergence. Extensive experiments have been conducted to evaluate the performance of the approach and more particularly of each component in the resolution of two classical combinatorial problems, the travelling salesman problem and quadratic assignment problem. Extensibility and quality of the solutions are analyzed compared to state-of-the-art parallel models.