Les modèles de Markov à sauts (appelés JMS pour Jump Markov System) sont utilisés dans de nombreux domaines tels que la poursuite de cibles, le traitement des signaux sismiques et la finance, étant donné leur bonne capacité à modéliser des systèmes non-linéaires et non-gaussiens. De nombreux travaux ont étudié les modèles de Markov linéaires pour lesquels bien souvent la restauration de données est réalisée grâce à des méthodes d’échantillonnage statistique de type Markov Chain Monte-Carlo. Dans cette thèse, nous avons cherché des solutions alternatives aux méthodes MCMC et proposons deux originalités principales. La première a consisté à proposer un algorithme de restauration non supervisée d’un JMS particulier appelé « modèle de Markov couple à sauts conditionnellement gaussiens » (noté CGPMSM). Cet algorithme combine une méthode d’estimation des paramètres basée sur le principe Espérance-Maximisation (EM) et une méthode efficace pour lisser les données à partir des paramètres estimés. La deuxième originalité a consisté à étendre un CGPMSM spécifique appelé CGOMSM par l’introduction des copules. Ce modèle, appelé GCOMSM, permet de considérer des distributions plus générales que les distributions gaussiennes tout en conservant des méthodes de restauration optimales et rapides. Nous avons équipé ce modèle d’une méthode d’estimation des paramètres appelée GICE-LS, combinant le principe de la méthode d’estimation conditionnelle itérative généralisée et le principe des moindre-carrés linéaires. Toutes les méthodes sont évaluées sur des données simulées. En particulier, les performances de GCOMSM sont discutées au regard de modèles de Markov non-linéaires et non-gaussiens tels que la volatilité stochastique, très utilisée dans le domaine de la finance. / Switching Markov Models, also called Jump Markov Systems (JMS), are widely used in many fields such as target tracking, seismic signal processing and finance, since they can approach non-Gaussian non-linear systems. A considerable amount of related work studies linear JMS in which data restoration is achieved by Markov Chain Monte-Carlo (MCMC) methods. In this dissertation, we try to find alternative restoration solution for JMS to MCMC methods. The main contribution of our work includes two parts. Firstly, an algorithm of unsupervised restoration for a recent linear JMS known as Conditionally Gaussian Pairwise Markov Switching Model (CGPMSM) is proposed. This algorithm combines a parameter estimation method named Double EM, which is based on the Expectation-Maximization (EM) principle applied twice sequentially, and an efficient approach for smoothing with estimated parameters. Secondly, we extend a specific sub-model of CGPMSM known as Conditionally Gaussian Observed Markov Switching Model (CGOMSM) to a more general one, named Generalized Conditionally Observed Markov Switching Model (GCOMSM) by introducing copulas. Comparing to CGOMSM, the proposed GCOMSM adopts inherently more flexible distributions and non-linear structures, while optimal restoration is feasible. In addition, an identification method called GICE-LS based on the Generalized Iterative Conditional Estimation (GICE) and the Least-Square (LS) principles is proposed for GCOMSM to approximate any non-Gaussian non-linear systems from their sample data set. All proposed methods are tested by simulation. Moreover, the performance of GCOMSM is discussed by application on other generable non-Gaussian non-linear Markov models, for example, on stochastic volatility models which are of great importance in finance.

Sistema de controle via rede (NCS) é um sistema de controle distribuído onde os sensores, atuadores e controladores estão alocados fisicamente em locais separados e são conectados através de uma rede de comunicação industrial. O NCS representa a evolução das arquiteturas de controle em rede, fornecendo maior modularidade e descentralização do controle, facilidade de diagnóstico e manutenção e menor custo. O desafio no desenvolvimento de um NCS é contornar os efeitos degenerativos causados por fatores que afetam o seu desempenho e estabilidade. Entre estes fatores estão o período de amostragem dos sinais, a perda de informações transmitidas na rede e os atrasos de comunicação. Buscando superar este desafio, este trabalho apresenta o desenvolvimento de NCS para aplicações em redes CAN baseado no uso da simulação e na proposta de uma estratégia de controle. A utilização de ferramentas de simulação de NCS, selecionadas através de um estudo comparativo e qualitativo, permitiu analisar o impacto de fatores degenerativos no desempenho de controle e estabilidade de NCS. Essa análise por simulação permitiu evidenciar o período de amostragem como o fator de maior influência para o projeto de NCS em redes CAN. Para superar o problema do período de amostragem, uma estratégia de controle adaptativo foi proposta. Essa estratégia usa informações de saída do NCS para automaticamente adaptar o período de amostragem das mensagens, garantindo desempenho de controle e diminuindo significativamente a ocupação da rede CAN. Experimentos realizados em uma plataforma de pesquisa sobre NCS demonstraram a confiabilidade e robustez do uso da estratégia de controle adaptativo, mesmo em condições extremas de operação da rede CAN. Os experimentos também permitiram comprovar a eficácia de uma técnica de identificação de NCS desenvolvida, que apresenta a vantagem de utilizar informações disponíveis na rede para obtenção de um modelo do NCS com precisão aceitável. / Networked control system (NCS) is a distributed control system where the sensors, actuators and controllers are physically separated and connected through an industrial communication network. The NCS represents the evolution of networked control architectures providing greater modularity and control decentralization, maintenance and diagnosis ease and lower cost of implementation. The challenge in the development of NCS is to overcome the degenerative effects of factors which affect its performance and stability. Among these factors are the sampling time, the loss of information on the network and the network delays. Aiming to overcome this challenge, this work presents the development of NCS for applications in CAN-Based networks based on the simulation use and in a control strategy proposal. The use NCS simulation tools, selected by a comparative and qualitative study, allowed to analyze the impact of degrading factors in the NCS control performance and stability. This analysis using simulation highlighted the message sampling time as factor with the biggest influence for the design of CAN-based NCS. To overcome the sampling time problem, an adaptive control strategy was proposed. This strategy uses the NCS output to automatically adapt the message sampling time, ensuring NCS control performance and stability and providing significant reduction of the CAN network load. Experiments carried out on a NCS Research Platform demonstrated the reliability and robustness of the adaptive control methodology application, even under worst case conditions of operation of the CAN-based network. Experiments have also proved the effectiveness of a model identification technique developed for NCS, which presents the advantage of using information available on the network to obtain the NCS model with acceptable accuracy.

Sistema de controle via rede (NCS) é um sistema de controle distribuído onde os sensores, atuadores e controladores estão alocados fisicamente em locais separados e são conectados através de uma rede de comunicação industrial. O NCS representa a evolução das arquiteturas de controle em rede, fornecendo maior modularidade e descentralização do controle, facilidade de diagnóstico e manutenção e menor custo. O desafio no desenvolvimento de um NCS é contornar os efeitos degenerativos causados por fatores que afetam o seu desempenho e estabilidade. Entre estes fatores estão o período de amostragem dos sinais, a perda de informações transmitidas na rede e os atrasos de comunicação. Buscando superar este desafio, este trabalho apresenta o desenvolvimento de NCS para aplicações em redes CAN baseado no uso da simulação e na proposta de uma estratégia de controle. A utilização de ferramentas de simulação de NCS, selecionadas através de um estudo comparativo e qualitativo, permitiu analisar o impacto de fatores degenerativos no desempenho de controle e estabilidade de NCS. Essa análise por simulação permitiu evidenciar o período de amostragem como o fator de maior influência para o projeto de NCS em redes CAN. Para superar o problema do período de amostragem, uma estratégia de controle adaptativo foi proposta. Essa estratégia usa informações de saída do NCS para automaticamente adaptar o período de amostragem das mensagens, garantindo desempenho de controle e diminuindo significativamente a ocupação da rede CAN. Experimentos realizados em uma plataforma de pesquisa sobre NCS demonstraram a confiabilidade e robustez do uso da estratégia de controle adaptativo, mesmo em condições extremas de operação da rede CAN. Os experimentos também permitiram comprovar a eficácia de uma técnica de identificação de NCS desenvolvida, que apresenta a vantagem de utilizar informações disponíveis na rede para obtenção de um modelo do NCS com precisão aceitável. / Networked control system (NCS) is a distributed control system where the sensors, actuators and controllers are physically separated and connected through an industrial communication network. The NCS represents the evolution of networked control architectures providing greater modularity and control decentralization, maintenance and diagnosis ease and lower cost of implementation. The challenge in the development of NCS is to overcome the degenerative effects of factors which affect its performance and stability. Among these factors are the sampling time, the loss of information on the network and the network delays. Aiming to overcome this challenge, this work presents the development of NCS for applications in CAN-Based networks based on the simulation use and in a control strategy proposal. The use NCS simulation tools, selected by a comparative and qualitative study, allowed to analyze the impact of degrading factors in the NCS control performance and stability. This analysis using simulation highlighted the message sampling time as factor with the biggest influence for the design of CAN-based NCS. To overcome the sampling time problem, an adaptive control strategy was proposed. This strategy uses the NCS output to automatically adapt the message sampling time, ensuring NCS control performance and stability and providing significant reduction of the CAN network load. Experiments carried out on a NCS Research Platform demonstrated the reliability and robustness of the adaptive control methodology application, even under worst case conditions of operation of the CAN-based network. Experiments have also proved the effectiveness of a model identification technique developed for NCS, which presents the advantage of using information available on the network to obtain the NCS model with acceptable accuracy.

La répression catabolique désigne un mode de régulation très répandu chez les bactéries, par lequel les enzymes nécessaires à l'import et la digestion de certaines sources carbonées sont réprimées en présence d'une source carbonée avantageuse, par exemple le glucose dans le cas de la bactérie E. coli. Nous proposons une approche mathématique et expérimentale pour séparer et évaluer l'importance des différents mécanismes de la répression catabolique. En particulier, nous montrons que l'AMP cyclique et l'état physiologique de la cellule jouent tous deux un rôle important dans la régulation de gènes sujets à la ré- pression catabolique. Nous présentons également des travaux méthodologiques réalisés dans le cadre de cette étude et contribuant à l'étude des réseaux de régulation génique en général. En particulier, nous étudions l'applicabilité de l'approximation quasi-stationnaire utilisée pour la réduction de modèles, et présentons des méthodes pour l'estimation robuste de taux de croissance, activité de promoteur, et concentration de protéines à partir de données bruitées provenant d'expériences avec gènes rapporteur. / Carbon Catabolite Repression (CCR) is a wide-spread mode of regulation in bacteria by which the enzymes necessary for the uptake and utilization of some carbon sources are repressed in presence of a preferred carbon source, e.g., glucose in the case of Escherichia coli . We propose a joint mathematical and experimental approach to separate and evaluate the importance of the different components of CCR. In particular, we show that both cyclic AMP and the global physiology of the cell play a major role in the regulation of the cAMP-dependent genes affected by CCR. We also present methodological improvements for the study of gene regulatory networks in general. In partic- ular, we examine the applicability of the Quasi-Steady-State-Approximation to reduce mathematical gene expression models, and provide robust meth- ods for the robust estimation of growth rate, promoter activity, and protein concentration from noisy kinetic reporter experiments.

[ES] En muchos campos de la ingeniería los modelos matemáticos son utilizados para describir el comportamiento de los sistemas, procesos o fenómenos. Hoy en día, existen varias técnicas o métodos que pueden ser usadas para obtener estos modelos. Debido a su versatilidad y simplicidad, a menudo se prefieren los métodos de identificación de sistemas. Por lo general, estos métodos requieren la definición de una estructura y la estimación computacional de los parámetros que la componen utilizando un conjunto de procedimientos y mediciones de las señales de entrada y salida del sistema. En el contexto de la identificación de sistemas no lineales, un desafío importante es la selección de la estructura. En el caso de que el sistema a identificar presente una no linealidad de tipo estático, los modelos orientados a bloques, pueden ser útiles para definir adecuadamente una estructura. Sin embargo, el diseñador puede enfrentarse a cierto grado de incertidumbre al seleccionar el modelo orientado a bloques adecuado en concordancia con el sistema real. Además de este inconveniente, se debe tener en cuenta que la estimación de algunos modelos orientados a bloques no es sencilla, como es el caso de los modelos de Wiener-Hammerstein que consisten en un bloque NL en medio de dos subsistemas LTI. La presencia de dos subsistemas LTI en los modelos de Wiener-Hammerstein es lo que principalmente dificulta su estimación. Generalmente, el procedimiento de identificación comienza con la estimación de la dinámica lineal, y el principal desafío es dividir esta dinámica entre los dos bloques LTI. Por lo general, esto implica una alta interacción del usuario para desarrollar varios procedimientos, y el modelo final estimado depende principalmente de estas etapas previas. El objetivo de esta tesis es contribuir a la identificación de los modelos de Wiener-Hammerstein. Esta contribución se basa en la presentación de dos nuevos algoritmos para atender aspectos específicos que no han sido abordados en la identificación de este tipo de modelos. El primer algoritmo, denominado WH-EA, permite estimar todos los parámetros de un modelo de Wiener-Hammerstein con un solo procedimiento a partir de un modelo dinámico lineal. Con WH-EA, una buena estimación no depende de procedimientos intermedios ya que el algoritmo evolutivo simultáneamente busca la mejor distribución de la dinámica, ajusta con precisión la ubicación de los polos y los ceros y captura la no linealidad estática. Otra ventaja importante de este algoritmo es que bajo consideraciones específicas y utilizando una señal de excitación adecuada, es posible crear un enfoque unificado que permite también la identificación de los modelos de Wiener y Hammerstein, que son casos particulares del modelo de Wiener-Hammerstein cuando uno de sus bloques LTI carece de dinámica. Lo interesante de este enfoque unificado es que con un mismo algoritmo es posible identificar los modelos de Wiener, Hammerstein y Wiener-Hammerstein sin que el usuario especifique de antemano el tipo de estructura a identificar. El segundo algoritmo llamado WH-MOEA, permite abordar el problema de identificación como un Problema de Optimización Multiobjetivo (MOOP). Sobre la base de este algoritmo se presenta un nuevo enfoque para la identificación de los modelos de Wiener-Hammerstein considerando un compromiso entre la precisión alcanzada y la complejidad del modelo. Con este enfoque es posible comparar varios modelos con diferentes prestaciones incluyendo como un objetivo de identificación el número de parámetros que puede tener el modelo estimado. El aporte de este enfoque se sustenta en el hecho de que en muchos problemas de ingeniería los requisitos de diseño y las preferencias del usuario no siempre apuntan a la precisión del modelo como un único objetivo, sino que muchas veces la complejidad es también un factor predominante en la toma de decisiones. / [CA] En molts camps de l'enginyeria els models matemàtics són utilitzats per a descriure el comportament dels sistemes, processos o fenòmens. Hui dia, existeixen diverses tècniques o mètodes que poden ser usades per a obtindre aquests models. A causa de la seua versatilitat i simplicitat, sovint es prefereixen els mètodes d'identificació de sistemes. En general, aquests mètodes requereixen la definició d'una estructura i l'estimació computacional dels paràmetres que la componen utilitzant un conjunt de procediments i mesuraments dels senyals d'entrada i eixida del sistema. En el context de la identificació de sistemes no lineals, un desafiament important és la selecció de l'estructura. En el cas que el sistema a identificar presente una no linealitat de tipus estàtic, els models orientats a blocs, poden ser útils per a definir adequadament una estructura. No obstant això, el dissenyador pot enfrontar-se a cert grau d'incertesa en seleccionar el model orientat a blocs adequat en concordança amb el sistema real. A més d'aquest inconvenient, s'ha de tindre en compte que l'estimació d'alguns models orientats a blocs no és senzilla, com és el cas dels models de Wiener-Hammerstein que consisteixen en un bloc NL enmig de dos subsistemes LTI. La presència de dos subsistemes LTI en els models de Wiener-Hammerstein és el que principalment dificulta la seua estimació. Generalment, el procediment d'identificació comença amb l'estimació de la dinàmica lineal, i el principal desafiament és dividir aquesta dinàmica entre els dos blocs LTI. En general, això implica una alta interacció de l'usuari per a desenvolupar diversos procediments, i el model final estimat depén principalment d'aquestes etapes prèvies. L'objectiu d'aquesta tesi és contribuir a la identificació dels models de Wiener-Hammerstein. Aquesta contribució es basa en la presentació de dos nous algorismes per a atendre aspectes específics que no han sigut adreçats en la identificació d'aquesta mena de models. El primer algorisme, denominat WH-EA (Algorisme Evolutiu per a la identificació de sistemes de Wiener-Hammerstein), permet estimar tots els paràmetres d'un model de Wiener-Hammerstein amb un sol procediment a partir d'un model dinàmic lineal. Amb WH-EA, una bona estimació no depén de procediments intermedis ja que l'algorisme evolutiu simultàniament busca la millor distribució de la dinàmica, afina la ubicació dels pols i els zeros i captura la no linealitat estàtica. Un altre avantatge important d'aquest algorisme és que sota consideracions específiques i utilitzant un senyal d'excitació adequada, és possible crear un enfocament unificat que permet també la identificació dels models de Wiener i Hammerstein, que són casos particulars del model de Wiener-Hammerstein quan un dels seus blocs LTI manca de dinàmica. L'interessant d'aquest enfocament unificat és que amb un mateix algorisme és possible identificar els models de Wiener, Hammerstein i Wiener-Hammerstein sense que l'usuari especifique per endavant el tipus d'estructura a identificar. El segon algorisme anomenat WH-MOEA (Algorisme evolutiu multi-objectiu per a la identificació de models de Wiener-Hammerstein), permet abordar el problema d'identificació com un Problema d'Optimització Multiobjectiu (MOOP). Sobre la base d'aquest algorisme es presenta un nou enfocament per a la identificació dels models de Wiener-Hammerstein considerant un compromís entre la precisió aconseguida i la complexitat del model. Amb aquest enfocament és possible comparar diversos models amb diferents prestacions incloent com un objectiu d'identificació el nombre de paràmetres que pot tindre el model estimat. L'aportació d'aquest enfocament se sustenta en el fet que en molts problemes d'enginyeria els requisits de disseny i les preferències de l'usuari no sempre apunten a la precisió del model com un únic objectiu, sinó que moltes vegades la complexitat és també un factor predominant en la presa de decisions. / [EN] In several engineering fields, mathematical models are used to describe the behaviour of systems, processes or phenomena. Nowadays, there are several techniques or methods for obtaining mathematical models. Because of their versatility and simplicity, system identification methods are often preferred. Generally, systems identification methods require defining a structure and estimating computationally the parameters that make it up, using a set of procedures y measurements of the system's input and output signals. In the context of nonlinear system identification, a significant challenge is the structure selection. In the case that the system to be identified presents a static type of nonlinearity, block-oriented models can be useful to define a suitable structure. However, the designer may face a certain degree of uncertainty when selecting the block-oriented model in accordance with the real system. In addition to this inconvenience, the estimation of some block-oriented models is not an easy task, as is the case with the Wiener-Hammerstein models consisting of a NL block in the middle of two LTI subsystems. The presence of two LTI subsystems in the Wiener-Hammerstein models is what mainly makes their estimation difficult. Generally, the identification procedure begins with the estimation of the linear dynamics, and the main challenge is to split this dynamic between the two LTI block. Usually, this implies a high user interaction to develop several procedures, and the final model estimated mostly depends on these previous stages. The aim of this thesis is to contribute to the identification of the Wiener-Hammerstein models. This contribution is based on the presentation of two new algorithms to address specific aspects that have not been addressed in the identification of this type of model. The first algorithm, called WH-EA (An Evolutionary Algorithm for Wiener-Hammerstein System Identification), allows estimating all the parameters of a Wiener-Hammerstein model with a single procedure from a linear dynamic model. With WH-EA, a good estimate does not depend on intermediate procedures since the evolutionary algorithm looks for the best dynamic division, while the locations of the poles and zeros are fine-tuned, and nonlinearity is captured simultaneously. Another significant advantage of this algorithm is that under specific considerations and using a suitable excitation signal; it is possible to create a unified approach that also allows the identification of Wiener and Hammerstein models which are particular cases of the Wiener-Hammerstein model when one of its LTI blocks lacks dynamics. What is interesting about this unified approach is that with the same algorithm, it is possible to identify Wiener, Hammerstein, and Wiener-Hammerstein models without the user specifying in advance the type of structure to be identified. The second algorithm called WH-MOEA (Multi-objective Evolutionary Algorithm for Wiener-Hammerstein identification), allows to address the identification problem as a Multi-Objective Optimisation Problem (MOOP). Based on this algorithm, a new approach for the identification of Wiener-Hammerstein models is presented considering a compromise between the accuracy achieved and the model complexity. With this approach, it is possible to compare several models with different performances, including as an identification target the number of parameters that the estimated model may have. The contribution of this approach is based on the fact that in many engineering problems the design requirements and user's preferences do not always point to the accuracy of the model as a single objective, but many times the complexity is also a predominant factor in decision-making. / Zambrano Abad, JC. (2021). Identification of nonlinear processes based on Wiener-Hammerstein models and heuristic optimization [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/171739 / TESIS

The main objective of the thesis was to investigate the feasibility of accurately and precisely fore- casting the number of both national and provincial bene ciaries of social security grants in South Africa, using simple autoregressive integrated moving average (ARIMA) models. The series of the monthly number of bene ciaries of the old age, child support, foster care and disability grants from April 2004 to March 2010 were used to achieve the objectives of the thesis. The conclusions from analysing the series were that: (1) ARIMA models for forecasting are province and grant-type spe- ci c; (2) for some grants, national forecasts obtained by aggregating provincial ARIMA forecasts are more accurate and precise than those obtained by ARIMA modelling national series; and (3) for some grants, forecasts obtained by modelling the latest half of the series were more accurate and precise than those obtained from modelling the full series. / Mathematical Sciences / M.Sc. (Statistics)

<p>Predmet istraživanja je identifikaciji i adaptivna estimacija<br />parametara široke klase linearnih sistema. Predloženi algoritmi<br />za adaptivnu estimaciju parametara su primenjivi na sisteme koji se<br />opisuju funkcijama prenosa proizvoljnog oblika, što uključuje sisteme<br />sa kašnjenjem, distribuiranim parametrima, frakcione sisteme i<br />druge sisteme opisane iracionalnim funkcijama prenosa. Na<br />posletku, dat je algoritam za identifikaciju CNG sistema koji se ne<br />izvršava u realnom vremenu i pretpostavlja da struktura modela nije<br />poznata unapred.</p> / <p>The subject of this research is the system identification and adaptive<br />parameter estimation in wide class of linear processes. Proposed<br />approaches for adaptive parameter estimation can be applied to systems<br />described by transfer functions of arbitrary form, including systems with<br />delay, distributed-paratemeter systems, fractional order systems, and other<br />system described by irrational transfer functions. In the final part, an offline<br />algorithm for identification of CNG system which does not assume any a<br />priori known model structure is proposed.</p>

The main objective of the thesis was to investigate the feasibility of accurately and precisely fore- casting the number of both national and provincial bene ciaries of social security grants in South Africa, using simple autoregressive integrated moving average (ARIMA) models. The series of the monthly number of bene ciaries of the old age, child support, foster care and disability grants from April 2004 to March 2010 were used to achieve the objectives of the thesis. The conclusions from analysing the series were that: (1) ARIMA models for forecasting are province and grant-type spe- ci c; (2) for some grants, national forecasts obtained by aggregating provincial ARIMA forecasts are more accurate and precise than those obtained by ARIMA modelling national series; and (3) for some grants, forecasts obtained by modelling the latest half of the series were more accurate and precise than those obtained from modelling the full series. / Mathematical Sciences / M.Sc. (Statistics)

Cette thèse est consacrée à l'étude de diagnostic de pannes pour les systèmes pile à combustible de type PEMFC. Le but est d'améliorer la fiabilité et la durabilité de la membrane électrolyte polymère afin de promouvoir la commercialisation de la technologie des piles à combustible. Les approches explorées dans cette thèse sont celles du diagnostic guidé par les données. Les techniques basées sur la reconnaissance de forme sont les plus utilisées. Dans ce travail, les variables considérées sont les tensions des cellules. Les résultats établis dans le cadre de la thèse peuvent être regroupés en trois contributions principales.La première contribution est constituée d'une étude comparative. Plus précisément, plusieurs méthodes sont explorées puis comparées en vue de déterminer une stratégie précise et offrant un coût de calcul optimal.La deuxième contribution concerne le diagnostic online sans connaissance complète des défauts au préalable. Il s'agit d'une technique adaptative qui permet d'appréhender l'apparition de nouveaux types de défauts. Cette technique est fondée sur la méthodologie SSM-SVM et les règles de détection et de localisation ont été améliorées pour répondre au problème du diagnostic en temps réel.La troisième contribution est obtenue à partir méthodologie fondée sur l'utilisation partielle de modèles dynamiques. Le principe de détection et localisation de défauts est fondé sur des techniques d'identification et sur la génération de résidus directement à partir des données d'exploitation.Toutes les stratégies proposées dans le cadre de la thèse ont été testées à travers des données expérimentales et validées sur un système embarqué. / Aiming at improving the reliability and durability of Polymer Electrolyte Membrane Fuel Cell (PEMFC) systems and promote the commercialization of fuel cell technologies, this thesis work is dedicated to the fault diagnosis study for PEMFC systems. Data-driven fault diagnosis is the main focus in this thesis. As a main branch of data-driven fault diagnosis, the methods based on pattern classification techniques are firstly studied. Taking individual fuel cell voltages as original diagnosis variables, several representative methodologies are investigated and compared from the perspective of online implementation.Specific to the defects of conventional classification based diagnosis methods, a novel diagnosis strategy is proposed. A new classifier named Sphere-Shaped Multi-class Support Vector Machine (SSM-SVM) and modified diagnostic rules are utilized to realize the novel fault recognition. While an incremental learning method is extended to achieve the online adaptation.Apart from the classification based diagnosis approach, a so-called partial model-based data-driven approach is introduced to handle PEMFC diagnosis in dynamic processes. With the aid of a subspace identification method (SIM), the model-based residual generation is designed directly from the normal and dynamic operating data. Then, fault detection and isolation are further realized by evaluating the generated residuals.The proposed diagnosis strategies have been verified using the experimental data which cover a set of representative faults and different PEMFC stacks. The preliminary online implementation results with an embedded system are also supplied.

Cette thèse s'inscrit dans le cadre du projet collaboratif MICROSOL, mené par Schneider Electric, et qui oeuvre pour le développement de micros centrales solaires thermodynamiques destinées à la production d'électricité en sites isolés (non connectés au réseau électrique) en exploitant l'énergie thermique du soleil. Le but de cette thèse étant le développement de lois de commande innovantes et efficaces pour la gestion de l'énergie de deux types de micros centrales solaires thermodynamiques : à base de moteur à cycle de Stirling et à base de machines à Cycle de Rankine Organique (ORC). Dans une première partie, nous considérons une centrale solaire thermodynamique à base de machine à cycle de Stirling hybridée à un supercondensateur comme moyen de stockage d'énergie tampon. Dans ce cadre, nous proposons une première loi de commande validée expérimentalement, associée au système de conversion d'énergie du moteur Stirling, qui dote le système de performances quasi optimales en termes de temps de réponse ce qui permet de réduire la taille du supercondensateur utilisé. Une deuxième loi de commande qui gère explicitement les contraintes du système tout en dotant ce dernier de performances optimales en terme de temps de réponse, est également proposée. Cette dernière loi de commande est en réalité plus qu'un simple contrôleur, elle constitue une méthodologie de contrôle applicable pour une famille de systèmes de conversion de l'énergie.Dans une deuxième partie, nous considérons une centrale solaire thermodynamique à base de machine à cycle de Rankine Organique (ORC) hybridée à un banc de batteries comme moyen de stockage d'énergie tampon. Etant donné que ce système fonctionne à vitesse de rotation fixe pour la génératrice asynchrone qui est connectée à un système de conversion d'énergie commercial, nous proposons une loi de commande prédictive qui agit sur la partie thermodynamique de ce système afin de le faire passer d'un point de fonctionnement à un autre, lors des appels de puissance des charges électriques, le plus rapidement possible (pour réduire le dimensionnement des batteries) tout en respectant les contraintes physiques du système. La loi de commande prédictive développée se base sur un modèle dynamique de la machine ORC identifié expérimentalement grâce à un algorithme d'identification nonlinéaire adéquat. / This Ph.D thesis was prepared in the scope of the MICROSOL project, ledby Schneider Electric, that aims at developing Off-grid solar thermodynamic micro powerplants exploiting the solar thermal energy. The aim of this thesis being the development of innovative and efficient control strategies for the energy management of two kinds of solar thermodynamic micro power plants: based on Stirling engine and based and Organic RankineCycle (ORC) machines.In a first part, we consider the Stirling based solar thermodynamic micro power planthybridized with a supercapacitor as an energy buffer. Within this framework, we propose afirst experimentally validated control strategy, associated to the energy conversion system ofthe Stirling engine, that endows the system with quasi optimal performances in term of settlingtime enabling the size reduction of the supercapacitor. A second control strategy that handlesexplicitly the system constraints while providing the system with optimal performances interm of settling time , is also proposed. This control strategy is in fact more than a simplecontroller, it is a control framework that holds for a family of energy conversion systems.In a second part, we consider the Organic Rankine Cycle (ORC) based thermodynamicmicro power plant hybridized with a battery bank as an energy buffer. Since this system worksat constant speed for the asynchronous generator electrically connected to a commercial energyconversion system, we propose a model predictive controller that acts on the thermodynamicpart of this system to move from an operating point to another, during the load power demandtransients, as fast as possible (to reduce the size of the battery banks) while respecting thephysical system constraints. The developed predictive controller is based upon a dynamicmodel, for the ORC power plant, identified experimentally thanks to an adequate nonlinearidentification algorithm.