Pour faciliter la transition énergétique vers la réduction de la consommation des énergies fossiles, la réduction des émissions de CO2 et l’intégration des sources d’énergie renouvelables, il convient d’étudier des stratégies permettant d’adapter en temps réel la gestion énergétique de manière optimale par rapport aux contraintes extérieures et intérieures du bâtiment. Des leviers d’action existent à l’échelle de l’îlot comme la mutualisation des productions et des consommations. Cette thèse présente l’application de techniques d’optimisation au développement de stratégies de gestion du chauffage électrique d’un bâtiment multizone comportant des logements et des bureaux. Pour réduire les temps de calcul par rapport à une approche de résolution globale d’un problème d’optimisation à grande échelle, les méthodes de décomposition-coordination ont été étudiées. Ces méthodes permettent de résoudre des sous-problèmes d’optimisation à l’échelle de chaque zone et de réintégrer les couplages entre zones (couplages thermiques et partage d’une ressource) avec une étape de coordination. Une méthode décomposée-coordonnée a été retenue pour chaque type de couplage permettant la mise en place d’une gestion en temps réel à l’échelle multizone. / To facilitate the energy transition towards the reduction of fossil fuels consumption, CO2 emissions and the integration of renewable energy sources, it is necessary to study realtime management strategies to adapt energy management in an optimal way according to external and internal perturbations and the evolution of the building. Solutions exist at the scale of blocks of buildings such as production and consumption pooling. This thesis presents the development of real-time management strategies, using optimisation techniques, for the electric heating of a multi-zone building mixing residential and tertiary uses. To decrease computation time compared to a global resolution approach of large-scale optimisation problems, decomposition-coordination methods were studied. These methods consist in solving sub-problems of optimisation in each zone and reintegrating the links between zones (thermal couplings and resource sharing) using a coordination step. One of these methods was chosen for each type of coupling allowing the implementation of real-time management at a multi-zone scale.

As colunas depropanizadoras existentes nas refinarias de petróleo têm como função a separação entre as correntes de propano e butano. O objetivo de controle nestas colunas é a especificação de um teor máximo de iso-butano e mais pesados (C4 +) na corrente de propano e do teor máximo de propano e mais leves (C3 -) na corrente de butano. Controladores multivariáveis tradicionais, que normalmente são implementados nas colunas depropanizadoras, apresentam grande dificuldade para manter os produtos dentro de suas especificações, isto se deve ao fato de que este processo apresenta um comportamento bastante não-linear ao longo de toda sua região de operação. Neste trabalho temos como objetivo estudar as dificuldades encontradas no projeto de controle para esse tipo de sistema e implantar na planta industrial um controlador multivariável utilizando múltiplos modelos para controle da coluna. Para realizarmos este estudo utilizamos o simulador de processos HYSYSÔ para verificarmos o comportamento estático e dinâmico do processo. Os modelos utilizados para representar o processo são aqueles obtidos durante o estudo do comportamento dinâmico. Para implantação do controlador na unidade industrial é utilizado o SICON (Sistema de Controle da Petrobras) sendo algumas de suas rotinas modificadas para permitir a inclusão dos múltiplos modelos. Durante o estudo são comparadas as performances dos controladores QDMC e MMPC (Multi-Model Predictive Control) resolvido através de um algoritmo para NLP (Non Linear Programming). O controlador multimodelos (MMPC) é apresentado na forma de variáveis de estado podendo controlar sistemas de grande porte, inclusive sistemas com dinâmicas lentas e rápidas. Esta formulação permite prever as variáveis controladas em instantes de tempo esparsos e diferentes para cada controlada. O MMPC é capaz de tratar problemas de controle não-linear usando modelos lineares, introduzindo o conceito de robustez com a utilização do conjunto de modelos. O MMPC exige um menor esforço de sintonia que o QDMC sendo adequado para uma região mais ampla de operação. / Depropanizer columns are used in oil refineries for the separation of the propane stream from the butane stream. The control objective of these columns is the specification of a maximum content of iso-butane and heavier components (C4+) in the propane product and the maximum content of propane and lighter components (C3-) in the butane roduct. Multivariable controllers usually mplemented in depropanizer columns frequently resent great difficulty to maintain the products inside their specification ranges. This deficiency is due to the fact that the process presents a quite non-linear behavior along its operating window. The objective of the present work is to study the difficulties found in the design of the control system for the aforesaid process, and to implement in an industrial plant a multivariable controller using multiple models for the control of the separation column. To accomplish this study we used the HYSYSÔ process simulator to verify the static and dynamic behavior of the process. The models used to represent the real process in the controller are those obtained during the study of the dynamic behavior. The controller implementation in the industrial unit was done with SICON (Control System of Petrobras), which had some of its routines modified to allow the inclusion of multiple models. Along the work, performances of QDMC and MMPC(Multi-Model Predictive Control) controllers were compared. MMPC was solved through an algorithm for NLP (Non Linear Programming). The Multi-Model (MMPC) controller was implemented using a state space formulation which allows for the implementation of very large systems and besides, systems with simultaneous slow and fast dynamics. This formulation allows to foresee the controlled variables at sparse sample instants, that can be distinct for each controlled variable. MMPC is able to handle non-linear control problems using linear models by introducing the robustness concept with the use of a set of models. MMPC demands a smaller tuning effort than QDMC, and can be adapted to a wide range of operating conditions.

Na implementação de controladores MPC, o desenvolvimento e a definição dos modelos do processo é a etapa mais crítica e a que mais consome tempo. Normalmente, os modelos são obtidos através de testes de identificação realizados na planta, onde se observam as respostas em malha aberta das variáveis controladas a perturbações introduzidas individualmente nas variáveis manipuladas. Por este motivo, a aplicação das técnicas de identificação em malha fechada a controladores MPC com restrições nas entradas e/ou saídas é, reconhecidamente, uma área de aplicação de interesse crescente. Neste trabalho é estudada a modificação do controlador MPC convencional através da inclusão de uma nova restrição de excitação em adição às restrições normais do controlador, com a finalidade de perturbar o processo de forma controlada, propiciando a identificação em malha fechada de modelos mais precisos do processo, a partir de modelos aproximados. São desenvolvidas quatro abordagens para implementação desta filosofia e apresentadas simulações para vários casos teóricos, utilizando modelos de dois processos industriais obtidos de artigos recentes relacionados a controle multivariável com incertezas nos modelos. Os resultados das simulações indicam que os dados produzidos permitiram a correta identificação dos modelos tanto no caso nominal (modelo igual à planta) quanto para casos onde a planta era diferente do modelo empregado para as predições do MPC. / In MPC implementation, the process models development and definition is the most critical and time consuming task. Normally, the models are obtained through plant identification tests where perturbations are individually introduced in the manipulated variable while the controlled variable open-loop behavior is observed. For this reason, the application of closed-loop identification techniques to MPC controllers with input or output constraints is a growing interest area. This work studies the traditional MPC controller modification with the inclusion of a new excitation constraint, in addition to input or output constraints, whose function is to perturb the process in a controlled way, permitting the closed-loop identification of more precise models, based on known approximated models. Four implementation methodologies are developed and some simulated theoretical cases are presented using models of two industrial processes extracted from recent papers related to multivariable control with models uncertainty. The simulation results show that the obtained datasets allow the identification of the correct model, both in the nominal case (when the model used by MPC is the true model of the plant) and in the uncertain case, where the model used by MPC is different from the true model.

Cette thèse a pour objectif la modélisation et la commande d'un robot parallèle actionné pneumatiquement destiné à des applications de prise et dépose d'objets. Les actionneurs pneumatiques sont des actionneurs à bas coût et ayant des rapports poids/puissance plus importants que les actionneurs électriques. Ceci a pour avantage de réduire le coût de revient du robot en augmentant sa capacité de charge. Cependant, du fait des fortes non linéarités qui les caractérisent (compressibilité de l'air, caractéristique de la valve, frottement, hystérésis ...), le principal obstacle à leur utilisation en robotique est leur commande de façon précise et robuste. C'est pour cela que plusieurs stratégies de commande ont été proposées et implémentées expérimentalement sur un banc d'essai utilisant trois types d'actionneurs pneumatiques: deux vérins et des muscles artificiels travaillant en mode antagoniste. Ces stratégies sont des schémas en cascade qui –après une linéarisant exacte obtenue sur la base du système non linéaire- combinent un contrôleur externe de position et une boucle interne de force (équivalente à la différence de pression) dans le cas des vérins pneumatiques. Pour les muscles artificiels, le même principe est utilisé sauf que la boucle interne de force est remplacée par une boucle qui régule le couple en contrôlant les pressions de chacun des deux muscles. Un contrôleur prédictif généralisé (GPC) est synthétisé pour la boucle de position permettant ainsi de réduire sensiblement les temps de réponse. Pour la boucle interne de pression, un contrôleur robuste multi-objectif combinant des performances H infinie avec des contraintes de placement de pôle a été synthétisé. L'utilisation d'inégalités matricielles affines (LMI) a permis de combiner les objectifs du contrôleur de façon très intuitive. Le choix de ce contrôleur robuste est motivé par la nécessité de rejeter les fortes variations de charge qui caractérisent les applications de prise-et-dépose d'objets. Deux autres stratégies pour la commande prédictive en effort ont été synthétisées sur les vérins et les muscles et ont donné des résultats très encourageants. Les résultats expérimentaux obtenus ont montré l'apport de ces lois de commande en termes de performances (réduction des temps de réponse, erreurs de suivi faibles) et de robustesse (bon rejet des perturbations). Une étude comparative des trois actionneurs testés a conduit au choix du vérin à double effet standard car offrant le meilleur compromis entre performances, robustesse et coût de l'actionneur. En se basant sur ce choix, un nouveau prototype de robot parallèle à deux degrés de liberté utilisant les vérins standards a été conçu, modélisé et commandé en utilisant les différentes stratégies en cascade proposées. L'implémentation expérimentale des algorithmes de commande a conduit à des résultats encourageants en termes de qualité de suivi et de rejet de perturbations. / The thesis objective is the control of a parallel robot driven with pneumatic actuators for pick-and-place applications. The advantage of using pneumatic actuators rather than electrical ones is that they are cheaper and have a bigger power-to-weight ratio which can increase the payload abilities of the robot. However, due to their strong nonlinearities such as air compressibility, valve characteristic, friction, and hysteresis, they are still difficult to control precisely and in a robust way.That is why the main contribution of the thesis is in the control area where different control schemes have been proposed and experimentally implemented on a test bench that involves three types of pneumatic actuators: two cylinders and agonist/antagonist artificial muscles. After the modeling and identification of the nonlinear models, different strategies have been developed: for cylinders, a cascade scheme which uses an outer position control loop and an inner force (or pressure difference) loop is used. For muscles, the inner force loop is replaced by a torque loop controlled by acting on the pressures in each muscle. For position, a Generalized Predictive Controller (GPC) is synthesized reducing sensibly the time responses. For the inner pressure loop, an LMI based multi objective controller is synthesized combining H infinity performances and pole placement constraints. The choice of a robust controller is motivated by the necessity of rejecting load variation disturban ces that characterize pick-and-place applications. On the other hand, two predictive control strategies with feedback linearization were implemented showing very encouraging results.The different experimental results have shown the interest of such strategies in terms of performances (time response reduction, good dynamic tracking) and robustness (disturbance rejection). The comparison of the three tested actuators led to the choice of the standard double acting cylinder because it offers the best compromise in terms of performances and cost. This cylinder was then used to design a planar parallel robot and the implementation of the proposed cascade strategies. The experimental tests showed encouraging results in terms of disturbance rejection and ability of tracking dynamic references and performing pick-and-place cycles.

The UK domestic sector accounts for more than a quarter of total energy use. This energy use can be reduced through more efficient building operations. The energy efficiency can be improved through better control of heating in houses, which account for a large portion of total energy consumption. The energy consumption can be lowered by using renewable energy systems, which will also help the UK government to meet its targets towards reduction in carbon emissions and generation of clean energy. Building control has gained considerable interest from researchers and much improved ways of control strategies for heating and hot water systems have been investigated. This intensified research is because heating systems represent a significant share of our primary energy consumption to meet thermal comfort and indoor air quality criteria. Advances in computing control and research in advanced control theory have made it possible to implement advanced controllers in building control applications. Heating control system is a difficult problem because of the non-linearities in the system and the wide range of operating conditions under which the system must function. A model of a two zone building was developed in this research to assess the performance of different control strategies. Two conventional (On-Off and proportional integral controllers) and one advanced control strategies (model predictive controller) were applied to a solar heating system combined with a heat pump. The building was modelled by using a lumped approach and different methods were deployed to obtain a suitable model for an air source heat pump. The control objectives were to reduce electricity costs by optimizing the operation of the heat pump, integrating the available solar energy, shifting electricity consumption to the cheaper night-time tariff and providing better thermal comfort to the occupants. Different climatic conditions were simulated to test the mentioned controllers. Both on-off and PI controllers were able to maintain the tank and room temperatures to the desired set-point temperatures however they did not make use of night-time electricity. PI controller and Model Predictive Controller (MPC) based on thermal comfort are developed in this thesis. Predicted mean vote (PMV) was used for controlling purposes and it was modelled by using room air and radiant temperatures as the varying parameters while assuming other parameters as constants. The MPC dealt well with the disturbances and occupancy patterns. Heat energy was also stored into the fabric by using lower night-time electricity tariffs. This research also investigated the issue of model mismatch and its effect on the prediction results of MPC. MPC performed well when there was no mismatch in the MPC model and simulation model but it struggled when there was a mismatch. A genetic algorithm (GA) known as a non-dominated sorting genetic algorithm (NSGA II) was used to solve two different objective functions, and the mixed objective from the application domain led to slightly superior results. Overall results showed that the MPC performed best by providing better thermal comfort, consuming less electric energy and making better use of cheap night-time electricity by load shifting and storing heat energy in the heating tank. The energy cost was reduced after using the model predictive controller.

A identificação de modelos é usualmente a tarefa mais significativa e demorada no trabalho de implementação e manutenção de sistemas de controle que usam Controle Preditivo baseado em Modelos (MPC) tendo em vista a complexidade da tarefa e a importância que o modelo possui para um bom desempenho do controlador. Após a implementação, o controlador tende a permanecer com o modelo original mesmo que mudanças de processo tenham ocorrido levando a uma degradação das ações do controlador. Este trabalho apresenta uma aplicação industrial de re-identificação em malha fechada. A metodologia de excitação da planta utilizada foi apresentada em Sotomayor et al. (2009). Tal técnica permite obter o comportamento das variáveis de processo sem desligar o MPC e sem modificar sua estrutura, aumentando assim, o automatismo e a segurança do procedimento de re-identificação. O sistema re-identificado foi uma coluna debutanizadora de uma refinaria brasileira sendo que os modelos fazem parte do controle preditivo multivariável dessa coluna de destilação. A metodologia foi aplicada com sucesso podendo-se obter os seis novos modelos para atualizar o controlador em questão, o que resultou em uma melhoria de seu desempenho. / Model identification is usually the most significant and time-consuming task of implementing and maintaining control systems based on models (MPC) concerning the complexity of the task and the importance of the model for a good performance of the controller. After being implemented the MPC tends to remain with the original model even after process changes have occurred, leading to a degradation of the controller actions. The present work shows an industrial application of closed-loop re-identification. The plant excitation methodology used here was presented in Sotomayor et al. (2009). Such technique allows for obtaining the behavior of the process variables with the MPC still working and without modifying the MPC structure, increasing automation and safety of the re-identification procedure. The system re-identified was a debutanizer column of a Brazilian refinery being the models part of the multivariable predictive control of this distillation column. The methodology was applied with reasonable success managing to obtain 6 new models to update this MPC, and resulting in improved control performance.

A identificação para controle é baseada especificamente na construção de modelos matemáticos a partir de dados experimentais, cuja finalidade é encontrar uma relação entre um conjunto de entradas e saídas de um processo dinâmico. Estes modelos são de fundamental importância para o projeto de controladores em processos industriais. No presente trabalho é realizada a identificação e o desenvolvimento do sistema de controle para uma planta piloto de neutralização de pH. O procedimento de identificação é baseado na coleta de dados reais do processo de neutralização de pH, operando em malha fechada. A estimativa dos modelos é realizada de duas formas: (1) estimar modelos que representem o comportamento de todo o sistema, incluindo os controladores PID do processo e (2) estimar modelos do processo com os dados coletados dos sinais de controle e as variáveis de saída do processo. Com os modelos do processo estimados projeta-se uma estratégia de controle MPC (Model Predictive Control), envolvendo dois esquemas de controle. O primeiro esquema calculará os set points ótimos que ingressarão nas malhas do processo. O segundo esquema calculará os sinais de controle ótimos que ingressarão diretamente no processo. O tipo de controlador MPC adotado é o QDMC (Quadratic Dynamic Matrix Control), permitindo restringir os sinais de entrada e saída do processo. A avaliação destes esquemas de controle é realizada mediante a mudança do set point das malhas do processo e a influência de perturbações. As perturbações são baseadas no aumento da vazão do ácido que ingressa no reator. / Identification for control system is based specifically on the mathematical models construction from experimental data, whose aim is to find a relationship between a set of inputs and outputs of a dynamic process. These models are fundamentally important for the industrial processes controllers design. In this work is performed the identification and development of the control system for a pH neutralization pilot plant. The identification procedure is based on the real data collected from pH neutralization process, operating in closed loop. The models estimation is performed in two forms: (1) estimating models that represent all system behavior, including process PID controllers and (2) estimating process models with collecting data of the control signals and process output variables. The process models parameters estimation is performed with the algorithms studied in Chapter 4. With the estimated process models is a MPC (Model Predictive Control) control strategy was designed, creating two control schemes. First scheme will compute the optimal set points that will enter to the process-loops. The second scheme will compute the optimal control signals that will enter to the process. The type of MPC controller adopted is a QDMC (Quadratic Dynamic Matrix Control), allowing restriction of the input and output signals. The control schemes evaluation is performed by changing the set point of the process-loops and the disturbance influence. This disturbance is based on acid flow increased that enters the reactor.

Colunas de destilação de alta pureza são sistemas de difícil controle. Apresentam longo tempo de resposta, comportamento altamente não linear e grande interação entre as variáveis. Os controladores preditivos são muito utilizados para o controle de colunas de destilação. Porém, em colunas de alta pureza, a incorporação de um único modelo linear geralmente acarreta em um controle de fraco desempenho. Isto pois, a representação do processo é deficiente, já que não considera variações nos ganhos e nas dinâmicas, típicas de um comportamento não linear. Estas incertezas podem, inclusive, provocar a instabilidade do controle o que resultaria em produtos que não atendam à especificação. Este trabalho tem por objetivo avaliar o desempenho dos controladores de horizonte de predição infinito com um modelo interno e com múltiplos modelos tendo o HYSYS(TM) como simulador de uma coluna de separação benzeno/tolueno e o MATLAB(TM) como ambiente para o controle supervisório. Observa-se que o controlador com apenas um modelo não é capaz de estabilizar o processo para perturbação nos valores de referência das variáveis controladas, ao contrário do controlador com múltiplos modelos. / High-purity distillation columns are systems which are typically difficult to control. The main reason for this is a strongly nonlinear and interactive system associated with a very sluggish response. Model Predictive Control is widely used for control of distillation columns. However, for high-purity columns, the use of a single linear model in the controller usually leads to a poor performance of the control system. The reason for this is the poor system representation, since variation in the system gains and time constants are not taken into account in the computation of the control law. Model uncertainties can produce instability in the control system and consequent deterioration of the product quality. The goal of this work is to evaluate the performance of infinite horizon MPC with a single internal model and with multiple models. HYSYS(TM) is used as simulator for the benzene/toluene column, and MATLAB(TM) is used as a platform for supervisory control. It is observed that the controller with a single model is not capable of stabilizing the process for disturbance in the set point of the controlled variables. Opposite to that behavior the controller with multiple models has a good performance.

In modern cyber-physical systems (CPSs) where the control signals are generally transmitted via shared communication networks, there is a desire to balance the closed-loop control performance with the communication cost necessary to achieve it. In this context, aperiodic real-time scheduling of control tasks comes into being and has received increasing attention recently. It is well known that model predictive control (MPC) is currently widely utilized in industrial control systems and has greatly increased profits in comparison with the proportional integral-derivative (PID) control. As communication and networks play more and more important roles in modern society, there is a great trend to upgrade and transform traditional industrial systems into CPSs, which naturally requires extending conventional MPC to communication-efficient MPC to save network resources. Motivated by this fact, we in this thesis propose robust MPC and scheduling co-design algorithms to networked CPSs possibly affected by both parameter uncertainties and additive disturbances. In Chapter 2, a dynamic event-triggered robust tube-based MPC for constrained linear systems with additive disturbances is developed, where a time-varying pre-stabilizing gain is obtained by interpolating multiple static state feedbacks and the interpolating coefficient is determined via optimization at the time instants when the MPC-based control is triggered. The original constraints are properly tightened to achieve robust constraint optimization and a sequence of dynamic sets used to test events are derived according to the optimized coefficient. We theoretically show that the proposed algorithm is recursively feasible and the closed-loop system is input-to-state stable (ISS) in the attraction region. Numerical results are presented to verify the design. In Chapter 3, a self-triggered min-max MPC strategy is developed for constrained nonlinear systems subject to both parametric uncertainties and additive disturbances, where the robust constraint satisfaction is achieved by considering the worst case of all possible uncertainty realizations. First, we propose a new cost function that relaxes the penalty on the system state in a time period where the controller will not be invoked. With this cost function, the next triggering time instant can be obtained at current time instant by solving a min-max optimization problem where the maximum triggering period becomes a decision variable. The proposed strategy is proved to be input-to-state practical stable (ISpS) in the attraction region at triggering time instants under some standard assumptions. Extensions are made to linear systems with additive disturbances, for which the conditions reduce to a linear matrix inequality (LMI). Comprehensive numerical experiments are performed to verify the correctness of the theoretical results. / Graduate

L’objectif de cette thèse est d’apporter des réponses à deux problèmes importants dans le domaine de l’automatique : l'estimation d'état et la commande prédictive robuste sous contraintes pour des systèmes incertains, en se basant sur des méthodes ensemblistes, plus précisément liées aux ensembles zonotopiques. Les incertitudes agissant sur le système sont modélisées de façon déterministe, elles sont donc inconnues mais bornées par des ensembles connus.Dans ce contexte, la première partie de la thèse développe une méthode d’estimation afin d’élaborer à chaque instant un ensemble zonotopique contenant l’état du système malgré la présence de perturbations, de bruits de mesure et d’incertitudes paramétriques définies par intervalle. Cette méthode est fondée sur la minimisation du P-rayon d’un zonotope, critère original permettant de caractériser la taille de l’ensemble zonotopique et réalisant un bon compromis entre la complexité et la précision de l’estimation. Cette approche est tout d’abord développée pour les systèmes mono-sortie, puis étendue au cas des systèmes multi-sorties, dans un premier temps par des extensions directes de la solution mono-sortie (le système multi-sorties est considéré comme plusieurs systèmes mono-sortie). Une autre solution est ensuite proposée, qui conduit à résoudre un problème d’optimisation de type Inégalités Matricielles Polynomiales en utilisant une méthode de relaxation. Les approches précédentes n’étant que des extensions de la solution à une seule sortie, et malgré leurs bons résultats obtenus en simulation, une démarche originale, dédiée aux systèmes multi-sorties, fondée sur l’intersection entre un polytope et un zonotope, est finalement développée et validée.La deuxième partie de la thèse aborde la problématique de la commande robuste par retour de sortie pour des systèmes incertains. La commande prédictive est retenue du fait de son utilisation dans de nombreux domaines, de sa facilité de mise en œuvre et de sa capacité à traiter des contraintes. Parmi les démarches issues de la littérature, l’implantation de techniques robustes fondées sur des tubes de trajectoire est développée plus spécifiquement. Le recours à un observateur ensembliste à base de zonotopes permet d’améliorer la qualité de l’estimation, ainsi que la performance de la commande, dans le cas de systèmes soumis à des perturbations et des bruits de mesure inconnus, mais bornés.Dans une dernière partie, cette combinaison de l’estimation ensembliste et de la commande prédictive robuste est testée en simulation sur un système de suspension magnétique. Les résultats de simulation traduisent un comportement tout à fait satisfaisant validant les structures théoriques élaborées. / The aim of this thesis is answering to two significant problems in the field of automatic control: the state estimation and the robust model predictive control for uncertain systems in the presence of input and state constraints, based on the set-membership approach, more precisely related to zonotopic sets. Uncertainties acting on the system are modeled via the deterministic approach, and thus they are unknown but bounded by a known set.In this context, the first part of the thesis proposes an estimation method to compute a zonotope containing the real states of the system, which are consistent with the disturbances, the measurement noise and the interval parametric uncertainties. This method is based on the minimization of the P-radius of a zonotope, which is an original criterion to characterize the size of the zonotope, in order to obtain a good trade-off between the complexity and the precision of the estimation. This approach is first developed for single-output systems, and then extended to the case of multi-output systems. The first solution for multi-output systems is a direct extension of the solution for single-output systems (the multi-output system being considered as several single-output systems). Another solution is then proposed, leading to solve a Polynomial Matrix Inequality optimization problem using a relaxation technique. Due to the fact that the previous approaches are just extensions of the solution for a single-output system, and despite their good performance results obtained in simulation, a novel approach dedicated to multi-output systems based on the intersection of a polytope and a zonotope is finally developed and validated.The second part of the thesis deals with the problem of robust output feedback control for uncertain systems. Model predictive control is chosen due to its use in many areas, its ability to deal with constraints and uncertainties. Among the approaches from the literature, the implementation of robust predictive techniques based on tubes of trajectories is developed. The use of a zonotopic set-membership estimation improves the quality of the estimation, as well as the performance of the control, for systems subject to unknown, but bounded disturbances and measurement noise.In the last part, the combination of zonotopic set-membership estimation and robust model predictive control is tested in simulation on a magnetic levitation system. The simulation results reflect a satisfactory behavior validating the developed theoretical techniques.