El texto completo de este trabajo no está disponible en el Repositorio Académico UPC por restricciones de la casa editorial donde ha sido publicado. / Protein fold recognition is an important task in the biological area. Different machine learning methods such as multiclass classifiers, one-vs-all and ensemble nested dichotomies were applied to this task and, in most of the cases, multiclass approaches were used. In this paper, we compare classifiers organized in tree structures to classify folds. We used a benchmark dataset containing 125 features to predict folds, comparing different supervised methods and achieving 54% of accuracy. An approach related to tree-structure of classifiers obtained better results in comparison with a hierarchical approach. / Revisión por pares

No description available.

Made available in DSpace on 2014-08-20T13:59:05Z (GMT). No. of bitstreams: 1 dissertacao_luis_willian_pacheco_arge.pdf: 2893589 bytes, checksum: d2e275d77781d5e9508058a77fa20f96 (MD5) Previous issue date: 2012-08-31 / This study aimed to assess the diversity, structure and the genetic relationship, evaluated by phenotypic and molecular of 75 Prunus rootstocks collection belonging to EMBRAPA Clima Temperado. The phenotypic analyzes were conducted by the evaluation of 21 qualitative and 26 quantitative traits of different plant organs and molecular analyzes were based on evaluation of 17 SSR loci. The data of quantitative traits were categorized by Scott & Knott method and submitted along with the qualitative data to statistical analysis (hierarchical by Jaccard coefficient) and clustering of phenotypic relationship. Molecular data were first converted to different formats and subjected to various statistical analyzes. The UPGMA dendogram obtained by genetic distance matrix calculated by Li & Nei coefficient, using the data of SSR loci evaluated was not able to distinguish between Tsukuba-1, Tsukuba-2 and Tsukuba-3 accesses, however, showed phenotypic effective to distinguish them. The dendograms of both analyzes together with the results of Bayesian approaches allowed the identification of three pools with high relation with the different groups that make up the collection. It was found that principal coordinates analysis based on phenotypic data, proved most effective for the detection of three pools detected with the hierarchical and the Bayesian approach. With the molecular data, the principal coordinates analysis corroborated with results obtained by the dendogram and the Bayesian approach. The access of group South Brazilian originating from samples collected in orchards in the region of Pelotas, which have no known pedigree, had largely down genetic and phenotypic distance, low 0.49 per both analyzes, with Aldrighi and Capdeboscq, and other known access. These accesses were traditionally used in the past as rootstocks in the state of Rio Grande do Sul. For both phenotypic and molecular analyzes, the group access of other species contributed more to genetic and phenotypic diversity, as expected, because are different species and with low similarity of features. Phenotypic and genetic characterization proved effective for elucidating the diversity, structure and genetic, and phenotypic relationship of the rootstocks of Prunus collection. / O presente trabalho objetivou avaliar a diversidade, estrutura e relação genética, por avaliação molecular, e fenotípica de uma coleção de 75 acessos de porta-enxertos de Prunus da EMBRAPA Clima Temperado. As análises fenotípicas foram conduzidas com a avaliação de 21 caracteres qualitativos e 26 quantitativos de diferentes órgãos das plantas, e as análises moleculares foram baseadas na avaliação de 17 loci SSR. Os dados dos caracteres quantitativos foram categorizados pelo método de agrupamento Scott & Knott e submetidos juntamente com os dados qualitativos às análises estatísticas de agrupamento (hierárquica pelo coeficiente de Jaccard) e de relação fenotípica. Os dados moleculares foram convertidos primeiramente para diferentes formatos e submetidos às diferentes análises estatísticas. O dendograma UPGMA obtido a partir da matriz de distância genética calculada pelo coeficiente Nei & Li, utilizando os dados dos loci SSR, não foi capaz de distinguir os acessos Tsukuba-1, Tsukuba-2 e Tsukuba-3, no entanto, os dados fenotípicos mostraram-se eficazes para distingui-los. Os dendograma de ambas as análises, juntamente com a abordagem Bayesiana, possibilitaram a identificação de três pools, com alta relação com os diferentes grupos que compõem a coleção. Verificou-se que análise de coordenadas principais, baseada em dados fenotípicos, mostrou-se mais eficaz para a detecção dos três pools detectados com as abordagens hierárquica e Bayesiana. Com os dados moleculares, a análise de coordenadas principais corroborou parcialmente com os resultados obtidos pelo dendograma e pela análise Bayesiana. Os acessos do grupo sul-brasileiro originários de coletas realizadas em pomares da região de Pelotas, os quais não possuem genealogia conhecida, apresentaram em grande parte baixa distância genética e fenotípica, abaixo de 0,49 por ambas as análises, em relação aos acessos Aldrighi e Capdeboscq. Estes acessos foram tradicionalmente usados no passado como porta-enxertos no estado do Rio Grande do Sul. Por ambas as análises, fenotípicas e moleculares, o grupo de acessos de outras espécies foi nalmente quem mais contribuiu para a diversidade genética e fenotípica, como já era esperado, pois são espécies distintas e com baixa similaridade de características. A caracterização genética por ambas as técnicas de análise mostrou-se efetiva para a elucidação da diversidade, estrutura e relação genética e fenotípica da coleção de porta-enxertos de Prunus avaliada.

Ce travail s’intéresse à la planification tactique de chaînes logistiques dans un environnement incertain et perturbé. Dans le cadre de relations « point-à point », nous proposons une approche permettant d’élaborer une planification tactique optimale et réactive d’un maillon d’une chaîne logistique en présence de paramètres incertains et de perturbations. Notre approche se fonde sur une structure à deux niveaux décisionnels. Le premier niveau effectue une planification agrégée en minimisant le coût global de production. Il établit ensuite « un plan de guidage » qui est transmis au niveau détaillé. Ce dernier effectue sa planification en suivant « au mieux » le plan de guidage et en prenant en compte les contraintes et données détaillées ignorées au niveau supérieur. Le niveau détaillé adopte un processus dynamique de planification à horizon glissant. Il réactualise ses données à chaque étape de planification afin d’assurer la réactivité du processus décisionnel. Nous caractérisons explicitement l’inertie du système décisionnel en distinguant deux phases : la phase d’anticipation et la phase de réalisation. Chaque phase est caractérisée par un délai temporel. Ainsi, nous proposons une modélisation originale du processus décisionnel de chaque décision via trois variables. Le niveau détaillé est formulé selon un programme linéaire.Au niveau agrégé, nous proposons un modèle global ayant l’originalité de prendre en compte les spécificités du processus décisionnel détaillé.Le couplage entre les deux niveaux est assuré par le plan de guidage. Selon les informations incluses dans le plan de guidage, le niveau agrégé accorde un certain degré d’autonomie au niveau détaillé, ceci conditionne la réactivité et la robustesse de la planification. Dans notre travail, nous considérons trois types de guidage : deux guidages budgétaires « globaux » et un guidage « prescriptif » par la sous-traitance agrégée.Notre approche est évaluée par simulation dans le cadre d’une demande incertaine. Pour cela, nous développons deux outils de simulation et un ensemble d’indicateurs de performances. Les expérimentations réalisées confirment la performance de notre approche par rapport à des approches classiques et mettent en évidence l’influence du type de guidage et du profil de la demande détaillée sur la réactivité et la robustesse des solutions trouvées. / This work focuses on the supply chain tactical planning problem in an uncertain and disrupted environment. As part of point-to-point relationships, we propose an optimal and reactive tactical planning approach of a supply chain link in the presence of uncertain parameters and disturbances.Our approach is based on a two-level decision structure. The first level performs an aggregate planning which minimizes the overall production cost. It establishes "a guiding plan" which is transmitted to the detailed level. This latter performs its planning by following a guiding plan and by taking into account detailed constraints and data.The detailed level adopts a dynamic planning process with a rolling horizon. It updates its data at every planning step to ensure a reactive decision making. We characterize explicitly the inertia of a decision making system by distinguishing two decision phases: the anticipation phase and the realization phase. Each phase is described by a time delay. Thus, we propose an original model of decision making process in which every decision is modeled by three variables. The detailed level is formulated according to a linear program.At the aggregate level, a view of the detailed decisional process is integrated by work-in-progress constraints. We propose an aggregate model whose originality is to consider the specifics of the detailed decision process.The coupling between the two levels is provided by the guiding plan. According to aggregated data included in this plan, the aggregate level gives a specific autonomy to the detailed level which conditions the reactivity and the robustness of the detailed planning. In our work, we consider three types of guidance: two “global” budget guidings and a more “precise” subcontracting aggregate guiding.Our approach is evaluated by simulation under uncertain demand. For this we develop two simulation tools and a set of performance indicators. The experiments carried out confirm the performance of our approach over conventional approaches and highlight the influence of the guiding and the detailed demand profile on the reactivity and robustness of the solutions

Ces travaux de thèse concernent l'analyse de la robustesse (stabilité et performance) de systèmes linéaires incertains de grande dimension avec une structure hiérarchique. Ces systèmes sont obtenus en interconnectant plusieurs sous-systèmes incertains à travers une topologie hiérarchique. L'analyse de la robustesse de ces systèmes est un problème à deux aspects : la robustesse et la grande dimension. La résolution efficace de ce problème en utilisant les approches usuelles est difficile, voire impossible, à cause de la complexité et de la grande taille du problème d'optimisation associé. La conséquence de cette complexité est une augmentation importante du temps de calcul nécessaire pour résoudre ce problème d'optimisation. Afin de réduire ce temps de calcul, les travaux existants ne considèrent que des classes particulières de systèmes linéaires incertains de grande dimension. De plus, la structure hiérarchique de ces systèmes n'est pas prise en compte, ce qui montre, de notre point de vue, les limitations de ces résultats. Notre objectif est d'exploiter la structure hiérarchique de ces systèmes afin de ramener la résolution du problème d'analyse de grande taille à la résolution d'un ensemble de problèmes d'analyse de faible taille, ce qui aura comme conséquence une diminution du temps de calcul. De plus, un autre avantage de cette approche est la possibilité de résoudre ces problèmes en même temps en utilisant le calcul parallèle. Afin de prendre en compte la structure hiérarchique du système incertain de grande dimension, nous modélisons ce dernier comme l'interconnexion de plusieurs sous-systèmes incertains qui sont eux-mêmes l'interconnexion d'autres sous-systèmes incertains, etc.. Cette technique récursive de modélisation est faite sur plusieurs niveaux hiérarchiques. Afin de réduire la complexité de la représentation des systèmes incertains, nous construisons une base de propriétés de dissipativité pour chaque sous-système incertain de chaque niveau hiérarchique. Cette base contient plusieurs éléments qui caractérisent des informations utiles sur le comportement de systèmes incertains. Des exemples de telles caractérisations sont : la caractérisation de la phase incertaine, la caractérisation du gain incertain, etc.. L'obtention de chaque élément est relaxée comme un problème d'optimisation convexe ou quasi-convexe sous contraintes LMI. L'analyse de la robustesse de systèmes incertains de grande dimension est ensuite faite de façon hiérarchique en propageant ces bases de propriétés de dissipativité d'un niveau hiérarchique à un autre. Nous proposons deux algorithmes d'analyse hiérarchique qui permettent de réduire le temps de calcul nécessaire pour analyser la robustesse de ces systèmes. Un avantage important de notre approche est la possibilité d'exécuter des parties de ces algorithmes de façon parallèle à chaque niveau hiérarchique ce qui diminuera de façon importante ce temps de calcul. Pour finir et dans le même contexte de système de grande dimension, nous nous intéressons à l'analyse de la performance dans les réseaux électriques et plus particulièrement «l'analyse du flux de puissances incertaines dans les réseaux électriques de distribution». Les sources d'énergies renouvelables comme les éoliennes et les panneaux solaires sont influencées par plusieurs facteurs : le vent, l'ensoleillement, etc.. Les puissances générées par ces sources sont alors intermittentes, variables et difficiles à prévoir. L'intégration de telles sources de puissance dans les réseaux électriques influencera les performances en introduisant des incertitudes sur les différentes tensions du réseau. L'analyse de l'impact des incertitudes de puissances sur les tensions est appelée «analyse du flux de puissances incertaines». La détermination de bornes sur les modules des différentes tensions est formulée comme un problème d'optimisation convexe sous contraintes LMI. / This PhD thesis concerns robustness analysis (stability and performance) of uncertain large scale systems with hierarchical structure. These systems are obtained by interconnecting several uncertain sub-systems through a hierarchical topology. Robustness analysis of these systems is a two aspect problem: robustness and large scale. The efficient resolution of this problem using usual approaches is difficult, even impossible, due to the high complexity and the large size of the associated optimization problem. The consequence of this complexity is an important increase of the computation time required to solve this optimization problem. In order to reduce this computation time, the existing results in the literature focus on particular classes of uncertain linear large scale systems. Furthermore, the hierarchical structure of the large scale system is not taken into account, which means, from our point of view, that these results have several limitations on different levels. Our objective is to exploit the hierarchical structure to obtain a set of small scale size optimization problems instead of one large scale optimization problem which will result in an important decrease in the computation time. Furthermore, another advantage of this approach is the possibility of solving these small scale optimization problems in the same time using parallel computing. In order to take into account the hierarchical structure, we model the uncertain large scale system as the interconnection of uncertain sub-systems which themselves are the interconnection of other uncertain sub-systems, etc.. This recursive modelling is performed at several hierarchical levels. In order to reduce the representation complexity of uncertain systems, we construct a basis of dissipativity properties for each uncertain sub-system at each hierarchical level. This basis contains several elements which characterize different useful information about uncertain system behaviour. Examples of such characterizations are: uncertain phase characterization, uncertain gain characterization, etc.. Obtaining each of these elements is relaxed as convex or quasi-convex optimization problem under LMI constraints. Robustness analysis of uncertain large scale systems is then performed in a hierarchical way by propagating these dissipativity property bases from one hierarchical level to another. We propose two hierarchical analysis algorithms which allow to reduce the computation time required to perform the robustness analysis of the large scale systems. Another key point of these algorithms is the possibility to be performed in parallel at each hierarchical level. The advantage of performing robustness analysis in parallel is an important decrease of the required computation time. Finally and within the same context of robustness analysis of uncertain large scale systems, we are interested in robustness analysis of power networks and more precisely in "the uncertain power flow analysis in distribution networks". The renewable energy resources such as solar panels and wind turbines are influenced by many factors: wind, solar irradiance, etc.. Therefore, the power generated by these resources is intermittent, variable and difficult to predict. The integration of such resources in power networks will influence the network performances by introducing uncertainties on the different network voltages. The analysis of the impact of power uncertainties on the voltages is called "uncertain power flow analysis". Obtaining the boundaries for the different modulus of these voltages is formulated as a convex optimization problem under LMI constraints