Orientador: Celso Pascoli Bottura / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-22T08:49:57Z (GMT). No. of bitstreams: 1 Giesbrecht_Mateus_D.pdf: 4188992 bytes, checksum: a2d91ff20132430d1389b8cd758b80bc (MD5) Previous issue date: 2012 / Resumo: Nesta tese é descrito como alguns problemas relacionados à identificação de sistemas discretos multivariáveis, à realização de séries temporais discretas multivariáveis e à modelagem de séries temporais discretas multivariáveis, podem ser formulados como problemas de otimização. Além da formulação dos problemas de otimização, nesta tese também são apresentadas algumas propostas imuno-inspiradas para a solução de cada um dos problemas, assim como os resultados e conclusões da aplicação dos métodos propostos. Os métodos aqui propostos apresentam resultados e performance melhores que aqueles obtidos por métodos conhecidos para solução dos problemas estudados, e podem ser aplicados em problemas cujas condições não sejam favoráveis para aplicação dos métodos conhecidos na literatura / Abstract: In this thesis it is described how some problems related to multivariable system identification, multivariable time series realization and multivariable time series modeling, can be formulated as optimization problems. Additionally, in this thesis some immune-inspired methods to solve each problem are also shown, and also the results and conclusions resultant from the application of the proposed methods. The performance and the results obtained with the methods here proposed are better than the results produced by known methods to solve the studied problems and can be applied even if the problem conditions are not suitable to the methods presented in the literature / Doutorado / Automação / Doutor em Engenharia Elétrica

Cette thèse porte sur l'estimation bayésienne d'état dans les séries temporelles modélisées à l'aide des variables latentes hybrides, c'est-à-dire dont la densité admet une composante discrète-finie et une composante continue. Des algorithmes généraux d'estimation des variables d'états dans les modèles de Markov partiellement observés à états hybrides sont proposés et comparés avec les méthodes de Monte-Carlo séquentielles sur un plan théorique et appliqué. Le résultat principal est que ces algorithmes permettent de réduire significativement le coût de calcul par rapport aux méthodes de Monte-Carlo séquentielles classiques / This thesis addresses the Bayesian estimation of hybrid-valued state variables in time series. The probability density function of a hybrid-valued random variable has a finite-discrete component and a continuous component. Diverse general algorithms for state estimation in partially observable Markov processesare introduced. These algorithms are compared with the sequential Monte-Carlo methods from a theoretical and a practical viewpoint. The main result is that the proposed methods require less processing time compared to the classic Monte-Carlo methods

This thesis deals with the identification and regulation of the "Ball & Plate" model. The thesis contains a description of the existing real model and the relevant mathematical and simulation model. The root hodograph method and the state space method are used to calculate the controller, especially the feedback controller with integrator. The final part of the work is devoted to the 3D model construction using Simulink and SimScape, which describes and visualizes the behavior of the real model and enables simulation experiments to be performed quickly and easily.

In this thesis the issue with the description, modeling and analysis of hybrid dynamic systems is solved. The main goal is to select appropriate method for solving hybrid dynamic systems and to aim to question on solving controllability, observability and stability of these systems. Two types of hybrid systems are covered in this work. In the first part, the description and simulation of hybrid systems that are compiled of continous and discrete parts with the systems of discrete affair are specified. For the description and modeling of hybrid automata and hybrid Petri nets have been used. The possibilities mentioned above are showned on the simple examples. Second part of this work covers the description, modeling and analysis of hybrid systems that are compiled of links continous and discreate dynamic system. The state space has been used. The universal state equations are mentioned and have been used for the description of the hybrid system and the structure of the matrixes A, B, C, D. The relation and basic method are spin off for the analysis of hybrid dynamic systems. From the analysis point of view, these are general methods for the controllability, observability, stability and steady state of hybrid dynamic systems. The examples of hybrid control systems for mentioned methods are specified. If it is allowed, the results are verified by the simulation in MATLAB Simulink background.

This thesis deals with complete design and manufacturing of autonomous two wheeled self-balancing robot. The goal of this thesis is to maintain the robot in up-right position and to follow black line using camera. The robot is controlled using Raspberry Pi and driven by DC motors. This thesis includes the design and implementation of hardware and software parts. Subsequently there was created the dynamic model in Matlab/Simulink. Based on this model, the LQR and PID controller was designed.

The thesis introduces an overview of techniques for filtering of unobserved variables using a state-space representation of a model and state inequality constraints. It is mainly aimed at a derivation of the linear Kalman filter, its extension into a form of a non-linear filter and imposing state constraints. The state uniform model with noise bounds and the sequential importance sampling, as a method of particle filters using Monte Carlo simulations, are described as alternative methods. These three methods are applied on a simple semi-structural model for a monetary policy analysis. The filtration is based on Czech macroeconomic data and reflects an imposed non-negative state constraint on the interest rate. Results of the algorithms are compared and discussed.

The dissertation consists of three papers which apply Bayesian econometric techniques to monitoring macroeconomic and macro-financial developments in the economy. Its aim is to illustrate how Bayesian methods can be employed in standard areas of economic research (estimating systemic risk in the banking sectors, nowcasting GDP growth) and also in a more original area (monitoring developments in sovereign bond markets). In the first essay, we address a task which analytical departments in central banks or commercial banks face very often - nowcasting foreign demand of a small open economy. On the example of the Czech economy, we propose an approach to nowcast foreign GDP growth rates for the Czech economy. For presentation purposes, we focus on three major trading partners: Germany, Slovakia and France. We opt for a simple method which is very general and which has proved successful in the literature: the method based on bridge equation models. A battery of models is evaluated based on a pseudo-real- time forecasting exercise. The results for Germany and France suggest that the models are more successful at backcasting, nowcasting and forecasting than the naive random walk benchmark model. At the same time, the various models considered are more or less successful depending on the forecast horizon....

This thesis aims to investigate the possibility of using a mathematical model to detect several common faults in a hydraulic rock drill. To this end, a parameterized state space model of the hydraulic drill, which simulate its behavior, is created. The model parameters are divided into two categories where different estimation methods are used to determine their values. The first category consists mainly of the parameters that are assumed to be invariant and independent of the various operating conditions. Experimental data are used to estimate these parameters. The other category is the variables that change depending on the machine’s current condition and operating settings. These include the response from the rock and internal leakages in the hydraulic drill. These parameters are estimated by integrating the impact piston position measurements in the simulation algorithm. The model is simulated for different fault modes, and the resulting estimated parameters are studied. It is shown that the resulting distributions for some of the estimated parameters differ between the fault modes, which makes fault detection possible. Furthermore, a condition monitoring system based on the estimated parameters provided by the model is designed and evaluated. It is shown that the performance and the robustness of the monitoring system depend on the machine’s operating settings and condition, where the system performs best for an operating pressure of 220 bar and the internal cylinder leakages.

Controlling a high-dimensional dynamical system with continuous state and action spaces in a partially unknown environment like a gas turbine is a challenging problem. So far often hard coded rules based on experts´ knowledge and experience are used. Machine learning techniques, which comprise the field of reinforcement learning, are generally only applied to sub-problems. A reason for this is that most standard RL approaches still fail to produce satisfactory results in those complex environments. Besides, they are rarely data-efficient, a fact which is crucial for most real-world applications, where the available amount of data is limited. In this thesis recurrent neural reinforcement learning approaches to identify and control dynamical systems in discrete time are presented. They form a novel connection between recurrent neural networks (RNN) and reinforcement learning (RL) techniques. RNN are used as they allow for the identification of dynamical systems in form of high-dimensional, non-linear state space models. Also, they have shown to be very data-efficient. In addition, a proof is given for their universal approximation capability of open dynamical systems. Moreover, it is pointed out that they are, in contrast to an often cited statement, well able to capture long-term dependencies. As a first step towards reinforcement learning, it is shown that RNN can well map and reconstruct (partially observable) MDP. In the so-called hybrid RNN approach, the resulting inner state of the network is then used as a basis for standard RL algorithms. The further developed recurrent control neural network combines system identification and determination of an optimal policy in one network. In contrast to most RL methods, it determines the optimal policy directly without making use of a value function. The methods are tested on several standard benchmark problems. In addition, they are applied to different kinds of gas turbine simulations of industrial scale.

Avec l'évolution des technologies, la complexité des systèmes informatiques ne cesse de s'accroître. Parmi ces systèmes, on retrouve les logiciels critiques qui doivent offrir une garantie de sûreté de fonctionnement qui s'avère crucial et pour lesquels un dysfonctionnement peut avoir des conséquences graves. Les méthodes formelles fournissent des outils permettant de garantir mathématiquement l'absence de certaines erreurs. Ces méthodes sont indispensables pour assurer les plus hauts niveaux de sûreté. Mais l'application de ces méthodes sur un code système bas niveau se heurte à des difficultés d'ordre pratique et théorique. Les principales difficultés concernent la prise en compte des aspects bas niveau, comme les pointeurs et les interactions avec le matériel spécifique. De plus, le fait que ces systèmes soient concurrents conduit à une augmentation exponentielle du nombre de comportements possibles, ce qui rend plus difficile leur vérification. Dans cette thèse, nous proposons une méthodologie pour la spécification et la vérification par model-checking de ce type de systèmes, en particulier, ceux implémentés en C. Cette méthodologie est basée sur la traduction de la sémantique de C en TLA+, un langage de spécification formel adapté à la modélisation de systèmes concurrents. Nous avons proposé un modèle de mémoire et d'exécution d'un programme C séquentiel en TLA+. En se basant sur ce modèle, nous avons proposé un ensemble de règles de traduction d'un code C en TLA+ que nous avons implémenté dans un outil, appelé C2TLA+. Nous avons montré comment ce modèle peut s'étendre pour modéliser les programmes C concurrents et gérer la synchronisation entre plusieurs processus ainsi que leur ordonnancement. Pour réduire la complexité du model-checking, nous avons proposé une technique permettant de réduire significativement la complexité de la vérification. Cette réduction consiste pour un code C à agglomérer une suite d'instructions lors de la génération du code TLA+, sous réserve d'un ensemble de conditions.Nous avons appliqué la méthodologie proposée dans cette thèse sur un cas d'étude réel issu de l'implémentation d'un micronoyau industriel,sur lequel nous avons vérifié un ensemble de propriétés fonctionnelles. L'application de la réduction a permis de réduire considérablement le temps de la vérification, ce qui la rend utilisable en pratique.Les résultats ont permis d'étudier le comportement du système, de vérifier certaines propriétés et de trouver des bugs indétectables par des simples tests. / Software systems are critical and complex. In order to guarantee their correctness, the use of formal methodsis important. These methods can be defined as mathematically based techniques, languages and tools for specifying and reasoning about systems. But, the application of formal methods to software systems, implemented in C, is challenging due to the presence of pointers, pointer arithmetic andinteraction with hardware. Moreover, software systems are often concurrent, making the verification process infeasible. This work provides a methodology to specify and verify C software systems usingmodel-checking technique. The proposed methodology is based on translating the semantics of Cinto TLA+, a formal specification language for reasoning about concurrent and reactive systems. We define a memory and execution model for a sequential program and a set of translation rules from C to TLA+ that we developed in a tool called C2TLA+. Based on this model, we show that it can be extended to support concurrency, synchronization primitives and process scheduling. Although model-checking is an efficient and automatic technique, it faces the state explosion problem when the system becomes large. To overcome this problem, we propose a state-space reduction technique. The latter is based on agglomerating a set of C instructions during the generation phase of the TLA+ specification. This methodology has been applied to a concrete case study, a microkernel of an industrial real-time operating system, on which a set of functional properties has been verified. The application of the agglomeration technique to the case study shows the usefulness of the proposed technique in reducing the complexity of verification. The obtained results allow us to study the behavior of the system and to find errors undetectable using traditional testing techniques.