[Truncated abstract. Please see the pdf version for the complete text. Also, formulae and special characters can only be approximated here. Please see the pdf version of this abstract for an accurate reproduction.] This thesis treats two aspects of intelligent control: The first part is about long-term optimization by approximating dynamic programming and in the second part a specific class of a fast neural network, related to support vector machines (SVMs), is considered. The first part relates to approximate dynamic programming, especially in the framework of adaptive critic designs (ACDs). Dynamic programming can be used to find an optimal decision or control policy over a long-term period. However, in practice it is difficult, and often impossible, to calculate a dynamic programming solution, due to the 'curse of dimensionality'. The adaptive critic design framework addresses this issue and tries to find a good solution by approximating the dynamic programming process for a stationary environment. In an adaptive critic design there are three modules, the plant or environment to be controlled, a critic to estimate the long-term cost and an action or controller module to produce the decision or control strategy. Even though there have been many publications on the subject over the past two decades, there are some points that have had less attention. While most of the publications address the training of the critic, one of the points that has not received systematic attention is training of the action module.¹ Normally, training starts with an arbitrary, hopefully stable, decision policy and its long-term cost is then estimated by the critic. Often the critic is a neural network that has to be trained, using a temporal difference and Bellman's principle of optimality. Once the critic network has converged, a policy improvement step is carried out by gradient descent to adjust the parameters of the controller network. Then the critic is retrained again to give the new long-term cost estimate. However, it would be preferable to focus more on extremal policies earlier in the training. Therefore, the Calculus of Variations is investigated to discard the idea of using the Euler equations to train the actor. However, an adaptive critic formulation for a continuous plant with a short-term cost as an integral cost density is made and the chain rule is applied to calculate the total derivative of the short-term cost with respect to the actor weights. This is different from the discrete systems, usually used in adaptive critics, which are used in conjunction with total ordered derivatives. This idea is then extended to second order derivatives such that Newton's method can be applied to speed up convergence. Based on this, an almost concurrent actor and critic training was proposed. The equations are developed for any non-linear system and short-term cost density function and these were tested on a linear quadratic regulator (LQR) setup. With this approach the solution to the actor and critic weights can be achieved in only a few actor-critic training cycles. Some other, more minor issues, in the adaptive critic framework are investigated, such as the influence of the discounting factor in the Bellman equation on total ordered derivatives, the target interpretation in backpropagation through time as moving and fixed targets, the relation between simultaneous recurrent networks and dynamic programming is stated and a reinterpretation of the recurrent generalized multilayer perceptron (GMLP) as a recurrent generalized finite impulse MLP (GFIR-MLP) is made. Another subject in this area that is investigated, is that of a hybrid dynamical system, characterized as a continuous plant and a set of basic feedback controllers, which are used to control the plant by finding a switching sequence to select one basic controller at a time. The special but important case is considered when the plant is linear but with some uncertainty in the state space and in the observation vector, and a quadratic cost function. This is a form of robust control, where a dynamic programming solution has to be calculated. &sup1Werbos comments that most treatment of action nets or policies either assume enumerative maximization, which is good only for small problems, except for the games of Backgammon or Go [1], or, gradient-based training. The latter is prone to difficulties with local minima due to the non-convex nature of the cost-to-go function. With incremental methods, such as backpropagation through time, calculus of variations and model-predictive control, the dangers of non-convexity of the cost-to-go function with respect to the control is much less than the with respect to the critic parameters, when the sampling times are small. Therefore, getting the critic right has priority. But with larger sampling times, when the control represents a more complex plan, non-convexity becomes more serious.

Thesis (MScEng)--Stellenbosch University, 2012. / ENGLISH ABSTRACT: Simultaneous localization and mapping (SLAM) is vital for autonomous robot navigation. The robot must build a map of its environment while tracking its own motion through that map. Although many solutions to this intricate problem have been proposed, one of the most prominent issues that still needs to be resolved is to accurately measure and track landmarks over time. In this thesis we investigate the use of stereo vision for this purpose. In order to find landmarks in images we explore the use of two feature detectors: the scale-invariant feature transform (SIFT) and speeded-up robust features (SURF). Both these algorithms find salient points in images and calculate a descriptor for each point that is invariant to scale, rotation and illumination. By using the descriptors we match these image features between stereo images and use the geometry of the system to calculate a set of 3D landmark measurements. A Taylor approximation of this transformation is used to derive a Gaussian noise model for the measurements. The measured landmarks are matched to landmarks in a map to find correspondences. We find that this process often incorrectly matches ambiguous landmarks. To find these mismatches we develop a novel outlier detection scheme based on the random sample consensus (RANSAC) framework. We use a similarity transformation for the RANSAC model and derive a probabilistic consensus measure that takes the uncertainties of landmark locations into account. Through simulation and practical tests we find that this method is a significant improvement on the standard approach of using the fundamental matrix. With accurately identified landmarks we are able to perform SLAM. We investigate the use of three popular SLAM algorithms: EKF SLAM, FastSLAM and FastSLAM 2. EKF SLAM uses a Gaussian distribution to describe the systems states and linearizes the motion and measurement equations with Taylor approximations. The two FastSLAM algorithms are based on the Rao-Blackwellized particle filter that uses particles to describe the robot states, and EKFs to estimate the landmark states. FastSLAM 2 uses a refinement process to decrease the size of the proposal distribution and in doing so decreases the number of particles needed for accurate SLAM. We test the three SLAM algorithms extensively in a simulation environment and find that all three are capable of very accurate results under the right circumstances. EKF SLAM displays extreme sensitivity to landmark mismatches. FastSLAM, on the other hand, is considerably more robust against landmark mismatches but is unable to describe the six-dimensional state vector required for 3D SLAM. FastSLAM 2 offers a good compromise between efficiency and accuracy, and performs well overall. In order to evaluate the complete system we test it with real world data. We find that our outlier detection algorithm is very effective and greatly increases the accuracy of the SLAM systems. We compare results obtained by all three SLAM systems, with both feature detection algorithms, against DGPS ground truth data and achieve accuracies comparable to other state-of-the-art systems. From our results we conclude that stereo vision is viable as a sensor for SLAM. / AFRIKAANSE OPSOMMING: Gelyktydige lokalisering en kartering (simultaneous localization and mapping, SLAM) is ’n noodsaaklike proses in outomatiese robot-navigasie. Die robot moet ’n kaart bou van sy omgewing en tegelykertyd sy eie beweging deur die kaart bepaal. Alhoewel daar baie oplossings vir hierdie ingewikkelde probleem bestaan, moet een belangrike saak nog opgelos word, naamlik om landmerke met verloop van tyd akkuraat op te spoor en te meet. In hierdie tesis ondersoek ons die moontlikheid om stereo-visie vir hierdie doel te gebruik. Ons ondersoek die gebruik van twee beeldkenmerk-onttrekkers: scale-invariant feature transform (SIFT) en speeded-up robust features (SURF). Altwee algoritmes vind toepaslike punte in beelde en bereken ’n beskrywer vir elke punt wat onveranderlik is ten opsigte van skaal, rotasie en beligting. Deur die beskrywer te gebruik, kan ons ooreenstemmende beeldkenmerke soek en die geometrie van die stelsel gebruik om ’n stel driedimensionele landmerkmetings te bereken. Ons gebruik ’n Taylor- benadering van hierdie transformasie om ’n Gaussiese ruis-model vir die metings te herlei. Die gemete landmerke se beskrywers word dan vergelyk met dié van landmerke in ’n kaart om ooreenkomste te vind. Hierdie proses maak egter dikwels foute. Om die foutiewe ooreenkomste op te spoor het ons ’n nuwe uitskieterherkenningsalgoritme ontwikkel wat gebaseer is op die RANSAC-raamwerk. Ons gebruik ’n gelykvormigheidstransformasie vir die RANSAC-model en lei ’n konsensusmate af wat die onsekerhede van die ligging van landmerke in ag neem. Met simulasie en praktiese toetse stel ons vas dat die metode ’n beduidende verbetering op die standaardprosedure, waar die fundamentele matriks gebruik word, is. Met ons akkuraat geïdentifiseerde landmerke kan ons dan SLAM uitvoer. Ons ondersoek die gebruik van drie SLAM-algoritmes: EKF SLAM, FastSLAM en FastSLAM 2. EKF SLAM gebruik ’n Gaussiese verspreiding om die stelseltoestande te beskryf en Taylor-benaderings om die bewegings- en meetvergelykings te lineariseer. Die twee FastSLAM-algoritmes is gebaseer op die Rao-Blackwell partikelfilter wat partikels gebruik om robottoestande te beskryf en EKF’s om die landmerktoestande af te skat. FastSLAM 2 gebruik ’n verfyningsproses om die grootte van die voorstelverspreiding te verminder en dus die aantal partikels wat vir akkurate SLAM benodig word, te verminder. Ons toets die drie SLAM-algoritmes deeglik in ’n simulasie-omgewing en vind dat al drie onder die regte omstandighede akkurate resultate kan behaal. EKF SLAM is egter baie sensitief vir foutiewe landmerkooreenkomste. FastSLAM is meer bestand daarteen, maar kan nie die sesdimensionele verspreiding wat vir 3D SLAM vereis word, beskryf nie. FastSLAM 2 bied ’n goeie kompromie tussen effektiwiteit en akkuraatheid, en presteer oor die algemeen goed. Ons toets die hele stelsel met werklike data om dit te evalueer, en vind dat ons uitskieterherkenningsalgoritme baie effektief is en die akkuraatheid van die SLAM-stelsels beduidend verbeter. Ons vergelyk resultate van die drie SLAM-stelsels met onafhanklike DGPS-data, wat as korrek beskou kan word, en behaal akkuraatheid wat vergelykbaar is met ander toonaangewende stelsels. Ons resultate lei tot die gevolgtrekking dat stereo-visie ’n lewensvatbare sensor vir SLAM is.

Orientador: Fernando Jose Von Zuben / Acompanha CD-ROM / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-13T02:48:36Z (GMT). No. of bitstreams: 1 Cazangi_RenatoReder_D.pdf: 8716830 bytes, checksum: 272657e08f1aeb5622ebeb4412c49048 (MD5) Previous issue date: 2008 / Resumo: Novas técnicas de navegação autônoma de robôs móveis visam suprir a crescente demanda pelo emprego de robôs em diversos setores da sociedade e junto a uma ampla gama de tarefas. Os desafios envolvidos no desenvolvimento do sistema que controla o robô permitem afirmar que a inteligência embarcada em robôs atuais ainda encontra-se em um nível incipiente e limitado. Neste trabalho, cinco frentes de pesquisa complementares são propostas visando estudar, teórica e praticamente, aspectos fundamentais de projeto e implementação de controladores autônomos inteligentes para robótica móvel. Metodologias de computação bio-inspirada e de robótica evolutiva provêem os conceitos e ferramentas que fundamentam as cinco propostas, as quais são validadas com base em sistemas de navegação concebidos e aplicados a problemas relevantes da área. Uma série de simulações computacionais em ambientes virtuais e experimentos com robôs reais é realizada, permitindo medir o alcance das contribuições e apontar as principais frentes de atuação que se abrem como perspectivas futuras da pesquisa. / Abstract: Novel techniques for autonomous robot navigation aim at fulfilling the growing demand for mobile robots in multiple segments of society and in a plethora of tasks. The challenges involved in developing the system which controls the robot allow to say that the intelligence embedded in the current robots is found to be still incipient and limited. In this work, five complementary research fronts are proposed intending to study, theoretical and practically, aspects which are fundamental to the design and implementation of intelligent autonomous controllers for mobile robotics. Bio-inspired computing and evolutionary robotics methodologies provide the concepts and tools underlying the five proposals, which are validated through navigation systems devised and applied to important problems. Numerous real robot experiments as well as computational simulations taking place in virtual environments are carried out, allowing for the evaluation of contributions and also the discussion of future possibilities. / Doutorado / Engenharia de Computação / Doutor em Engenharia Elétrica

Le développement de véhicules autonomes capables de se déplacer sur les routes urbaines peuvent fournir des avantages importants en matière de réduction des accidents, en augmentant le confort et aussi, permettant des réductions de coûts. Les véhicules Intelligents par exemple fondent souvent leurs décisions sur les observations obtenues à partir de différents capteurs tels que les LIDAR, les GPS et les Caméras. En fait, les capteurs de la caméra ont reçu grande attention en raison du fait de qu’ils ne sont pas cher, facile à utiliser et fournissent des données avec de riches informations. Les environnements urbains représentent des scénarios intéressant mais aussi très difficile dans ce contexte, où le tracé de la route peut être très complexe,la présence d’objets tels que des arbres, des vélos, des voitures peuvent générer des observations partielles et aussi ces observations sont souvent bruyants ou même manquant en raison de occlusions complètes. Donc, le processus de perception par nature doit être capable de traiter des incertitudes dans la connaissance du monde autour de la voiture. Tandis que la navigation routière et la conduite autonome en utilisant une connaissance préalable de l’environnement ont démontré avec succès, la compréhension et la navigation des scénarios généraux du environnement urbain avec peu de connaissances reste un problème non résolu. Dans cette thèse, on analyse ce problème de perception pour la conduite dans les milieux urbains basée sur la connaissance de l’environnement pour aussi prendre des décisions dans la navigation autonome. Il est conçu un système de perception robotique, qui permettre aux voitures de se conduire sur les routes, sans la nécessité d’adapter l’infrastructure, sans exiger l’apprentissage précédente de l’environnement, et en tenant en compte la présence d’objets dynamiques tels que les voitures.On propose un nouveau procédé basé sur l’apprentissage par la machine pour extraire le contexte sémantique en utilisant une paire d’images stéréo qui est fusionnée dans une grille d’occupation évidentielle pour modéliser les incertitudes d’un environnement urbain inconnu,en utilisant la théorie de Dempster-Shafer. Pour prendre des décisions dans la planification des chemins, il est appliqué l’approche de tentacule virtuel pour générer les possibles chemins à partir du centre de référence de la voiture et sur cette base, deux nouvelles stratégies sont proposées. Première, une nouvelle stratégie pour sélectionner le chemin correct pour mieux éviter les obstacles et de suivre la tâche locale dans le contexte de la navigation hybride, et seconde, un nouveau contrôle en boucle fermée basé sur l’odométrie visuelle et tentacule virtuel est modélisée pour l’exécution du suivi de chemin. Finalement, un système complet automobile intégrant les modules de perception, de planification et de contrôle sont mis en place et validé expérimentalement dans des situations réelles en utilisant une voiture autonome expérimentale, où les résultats montrent que l’approche développée effectue avec succès une navigation locale fiable basée sur des capteurs de la caméra. / The development of autonomous vehicles capable of getting around on urban roads can provide important benefits in reducing accidents, in increasing life comfort and also in providing cost savings. Intelligent vehicles for example often base their decisions on observations obtained from various sensors such as LIDAR, GPS and Cameras. Actually, camera sensors have been receiving large attention due to they are cheap, easy to employ and provide rich data information. Inner-city environments represent an interesting but also very challenging scenario in this context,where the road layout may be very complex, the presence of objects such as trees, bicycles,cars might generate partial observations and also these observations are often noisy or even missing due to heavy occlusions. Thus, perception process by nature needs to be able to dea lwith uncertainties in the knowledge of the world around the car. While highway navigation and autonomous driving using a prior knowledge of the environment have been demonstrating successfully,understanding and navigating general inner-city scenarios with little prior knowledge remains an unsolved problem. In this thesis, this perception problem is analyzed for driving in the inner-city environments associated with the capacity to perform a safe displacement basedon decision-making process in autonomous navigation. It is designed a perception system that allows robotic-cars to drive autonomously on roads, with out the need to adapt the infrastructure,without requiring previous knowledge of the environment and considering the presenceof dynamic objects such as cars. It is proposed a novel method based on machine learning to extract the semantic context using a pair of stereo images, which is merged in an evidential grid to model the uncertainties of an unknown urban environment, applying the Dempster-Shafer theory. To make decisions in path-planning, it is applied the virtual tentacle approach to generate possible paths starting from ego-referenced car and based on it, two news strategies are proposed. First one, a new strategy to select the correct path to better avoid obstacles and tofollow the local task in the context of hybrid navigation, and second, a new closed loop control based on visual odometry and virtual tentacle is modeled to path-following execution. Finally, a complete automotive system integrating the perception, path-planning and control modules are implemented and experimentally validated in real situations using an experimental autonomous car, where the results show that the developed approach successfully performs a safe local navigation based on camera sensors.

Orientador: Fernando José Von Zuben / Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação / Made available in DSpace on 2018-08-18T14:21:13Z (GMT). No. of bitstreams: 1 Azzolini_AlissonGusatti_M.pdf: 3567259 bytes, checksum: 633eb00350cdfa625d0e628fdf1f247e (MD5) Previous issue date: 2011 / Resumo: Há uma demanda crescente por soluções avançadas de navegação autônoma em robótica móvel. Apresenta-se então um sistema de síntese e aprendizagem de controladores com tal finalidade. Propõe-se um controlador probabilístico, consistindo no acoplamento de um processo de decisão de Markov parcialmente observável (POMDP) com um classificador logístico multinomial. A parametrização empregada para o POMDP inspira-se numa proposta anterior de controle de robô por meio de redes imunológicas artificiais, que mostrou apresentar flexibilidade e capacidade de representação de conhecimento na execução de tarefas desafiadoras de navegação autônoma. A aprendizagem dos parâmetros do classificador logístico é efetuada através de um algoritmo de aprendizagem por reforço baseado em gradiente de política, e os do POMDP, atráves de um algoritmo de maximização de verossimilhança. Três experimentos computacionais são efetuados, dois deles utilizando somente o classificador logístico, e o terceiro utilizando o acoplamento entre POMDP e classificador logístico. Os resultados permitem a constatação de pontos fortes e algumas deficiências das duas abordagens. O trabalho aponta também para uma potencial reinterpretação do controlador baseado em rede imunológica em termos de um modelo probabilístico similar ao proposto / Abstract: There is an increasing demand for advanced solutions in autonomous navigation of mobile robots. A system is presented for the synthesis and learning of controllers for such purpose. A probabilistic controller is proposed, consisting of the coupling of a partially observable Markov decision process (POMDP) with a multinomial logistic classifier. The parametrization used for the POMDP draws on an earlier proposal of robot control based on artificial immune networks, that has shown to present flexibility and knowledge representation capability in the execution of challenging autonomous navigation tasks. Learning the logistic classifier parameters is accomplished through a reinforcement learning algorithm based on policy gradient, while the POMDP parameters are learned by a likelihood maximization algorithm. Three computational experiments are performed, two of them using only the logistic classifier, and the third one using the coupling of a POMDP with a logistic classifier. The results show some strong points and drawbacks of both approaches. The work also points torwards a potential reinterpretation of the immune network based controller in terms of a probabilistic model similar to the one proposed / Mestrado / Engenharia de Computação / Mestre em Engenharia Elétrica

Orientadores: Janito Vaqueiro Ferreira, Alessandro Corrêa Victorino / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Mecânica / Made available in DSpace on 2018-08-26T17:57:25Z (GMT). No. of bitstreams: 1 Vitor_GiovaniBernardes_D.pdf: 28262004 bytes, checksum: eeccacc4c01faa822412782af2e96121 (MD5) Previous issue date: 2014 / Resumo: O desenvolvimento de veículos autônomos capazes de se locomover em ruas urbanas pode proporcionar importantes benefícios na redução de acidentes, no aumentando da qualidade de vida e também na redução de custos. Veículos inteligentes, por exemplo, frequentemente baseiam suas decisões em observações obtidas a partir de vários sensores tais como LIDAR, GPS e câmeras. Atualmente, sensores de câmera têm recebido grande atenção pelo motivo de que eles são de baixo custo, fáceis de utilizar e fornecem dados com rica informação. Ambientes urbanos representam um interessante mas também desafiador cenário neste contexto, onde o traçado das ruas podem ser muito complexos, a presença de objetos tais como árvores, bicicletas, veículos podem gerar observações parciais e também estas observações são muitas vezes ruidosas ou ainda perdidas devido a completas oclusões. Portanto, o processo de percepção por natureza precisa ser capaz de lidar com a incerteza no conhecimento do mundo em torno do veículo. Nesta tese, este problema de percepção é analisado para a condução nos ambientes urbanos associado com a capacidade de realizar um deslocamento seguro baseado no processo de tomada de decisão em navegação autônoma. Projeta-se um sistema de percepção que permita veículos robóticos a trafegar autonomamente nas ruas, sem a necessidade de adaptar a infraestrutura, sem o conhecimento prévio do ambiente e considerando a presença de objetos dinâmicos tais como veículos. Propõe-se um novo método baseado em aprendizado de máquina para extrair o contexto semântico usando um par de imagens estéreo, a qual é vinculada a uma grade de ocupação evidencial que modela as incertezas de um ambiente urbano desconhecido, aplicando a teoria de Dempster-Shafer. Para a tomada de decisão no planejamento do caminho, aplica-se a abordagem dos tentáculos virtuais para gerar possíveis caminhos a partir do centro de referencia do veículo e com base nisto, duas novas estratégias são propostas. Em primeiro, uma nova estratégia para escolher o caminho correto para melhor evitar obstáculos e seguir a tarefa local no contexto da navegação hibrida e, em segundo, um novo controle de malha fechada baseado na odometria visual e o tentáculo virtual é modelado para execução do seguimento de caminho. Finalmente, um completo sistema automotivo integrando os modelos de percepção, planejamento e controle são implementados e validados experimentalmente em condições reais usando um veículo autônomo experimental, onde os resultados mostram que a abordagem desenvolvida realiza com sucesso uma segura navegação local com base em sensores de câmera / Abstract: The development of autonomous vehicles capable of getting around on urban roads can provide important benefits in reducing accidents, in increasing life comfort and also in providing cost savings. Intelligent vehicles for example often base their decisions on observations obtained from various sensors such as LIDAR, GPS and Cameras. Actually, camera sensors have been receiving large attention due to they are cheap, easy to employ and provide rich data information. Inner-city environments represent an interesting but also very challenging scenario in this context, where the road layout may be very complex, the presence of objects such as trees, bicycles, cars might generate partial observations and also these observations are often noisy or even missing due to heavy occlusions. Thus, perception process by nature needs to be able to deal with uncertainties in the knowledge of the world around the car. While highway navigation and autonomous driving using a prior knowledge of the environment have been demonstrating successfully, understanding and navigating general inner-city scenarios with little prior knowledge remains an unsolved problem. In this thesis, this perception problem is analyzed for driving in the inner-city environments associated with the capacity to perform a safe displacement based on decision-making process in autonomous navigation. It is designed a perception system that allows robotic-cars to drive autonomously on roads, without the need to adapt the infrastructure, without requiring previous knowledge of the environment and considering the presence of dynamic objects such as cars. It is proposed a novel method based on machine learning to extract the semantic context using a pair of stereo images, which is merged in an evidential grid to model the uncertainties of an unknown urban environment, applying the Dempster-Shafer theory. To make decisions in path-planning, it is applied the virtual tentacle approach to generate possible paths starting from ego-referenced car and based on it, two news strategies are proposed. First one, a new strategy to select the correct path to better avoid obstacles and to follow the local task in the context of hybrid navigation, and second, a new closed loop control based on visual odometry and virtual tentacle is modeled to path-following execution. Finally, a complete automotive system integrating the perception, path-planning and control modules are implemented and experimentally validated in real situations using an experimental autonomous car, where the results show that the developed approach successfully performs a safe local navigation based on camera sensors / Doutorado / Mecanica dos Sólidos e Projeto Mecanico / Doutor em Engenharia Mecânica

A variety of factors such as increasing electrical energy demand, slow expansion of transmission infrastructures, and electric energy market deregulation, are forcing utilities and system operators to operate power systems closer to their design limits. Operating under stressed regimes can have a detrimental effect on the rotor-angle stability of the system. This stability reduction is often reflected by the emergence or worsening of poorly damped low-frequency electromechanical oscillations. Without appropriate measures these can lead to costly blackouts. To guarantee system security, operators are sometimes forced to limit power transfers that are economically beneficial but that can result in poorly damped oscillations. Controllers that damp these oscillations can improve system reliability by preventing blackouts and provide long term economic gains by enabling more extensive utilization of the transmission infrastructure. Previous research in the use of artificial neural network-based intelligent controllers for power system damping control has shown promise when tested in small power system models. However, these controllers do not scale-up well enough to be deployed in realistically-sized power systems. The work in this dissertation focuses on improving the scalability of intelligent power system stabilizing controls so that they can significantly improve the rotor-angle stability of large-scale power systems. A framework for designing effective and robust intelligent controllers capable of scaling-up to large scale power systems is proposed. Extensive simulation results on a large-scale power system simulation model demonstrate the rotor-angle stability improvements attained by controllers designed using this framework.

166f. / Submitted by Suelen Reis (suziy.ellen@gmail.com) on 2013-04-10T19:35:07Z No. of bitstreams: 1 Luciana Pacheco seg.pdf: 3529890 bytes, checksum: 13857ac04543f1bbc9fd4d7ed9849eba (MD5) / Approved for entry into archive by Rodrigo Meirelles(rodrigomei@ufba.br) on 2013-05-11T15:30:28Z (GMT) No. of bitstreams: 1 Luciana Pacheco seg.pdf: 3529890 bytes, checksum: 13857ac04543f1bbc9fd4d7ed9849eba (MD5) / Made available in DSpace on 2013-05-11T15:30:28Z (GMT). No. of bitstreams: 1 Luciana Pacheco seg.pdf: 3529890 bytes, checksum: 13857ac04543f1bbc9fd4d7ed9849eba (MD5) Previous issue date: 2011 / Em sistemas industriais automatizados, a inatividade provocada pela escassez não planejada de recursos, ou por falhas de processo, tem grande influência no desempenho dos sistemas por conta das descontinuidades e instabilidades geradas. Sistemas de controle distribuídos e autônomos podem ajudar a lidar com esses tipos de problemas devido à melhoria de desempenho possibilitada. Entretanto, aspectos relativos à segurança e ao tempo de resposta devem ser bem tratados nesses sistemas devido aos riscos envolvidos (humanos, financeiros e ambientais). A proposta de sistemas autônomos e distribuídos visa a que decisões de controle sejam tomadas mais próximas do objeto controlado, reduzindo assim o tempo de atuação no processo e sistematizando algumas decisões, antes tomadas de forma empírica. Consequentemente, se espera aumentar a disponibilidade e a continuidade do processo, bem como garantir os aspectos de confiabilidade. Entretanto, quando tais sistemas se tornam mais autônomos e distribuídos, podem tender ao comportamento global caótico, caso suas interações não estejam bem definidas. Assim, é importante que seja avaliado e dimensionado o acoplamento entre os sistemas autônomos relacionados. O grau de inteligência de um sistema pode variar de uma entidade completamente controlada a entidades completamente autônomas. O primeiro nível de inteligência é verificado quando um sistema é capaz de gerenciar suas próprias informações, obtidas por meio de sensores e demais técnicas e dispositivos, e não somente manipular informações. Em um segundo nível, o sistema pode notificar o seu gestor quando há um problema. Em um terceiro nível, o sistema já é capaz de tomar decisões e se autogerenciar, mesmo sem intervenção externa. Neste caso, o sistema tem controle total sobre suas tarefas e não há nenhum controle externo a ele. A alternativa proposta pelo GCAD visa a que Sistemas Industriais Automatizados atinjam até o terceiro nível de inteligência, sendo que intervenções externas podem ser admitidas nos casos em que uma ação puramente local e autônoma de fato não é recomendável ou não é possível, por exemplo, havendo necessidade de substituição de equipamentos ou dispositivos. O GCAD propõe um módulo de controle inteligente instanciado predominantemente em nível local que visa a permitir que cada Sistema Industrial Automatizado, distribuído em células, tome decisões críticas de uma forma autônoma. Adicionalmente, um módulo remoto deve gerenciar situações mais complexas que estão além da capacidade de decisão ou atuação do sistema de controle local. O modelo proposto visa a permitir ajustes automáticos e autônomos no sistema, a fim de melhorar seu desempenho, e prevenir ou tratar as falhas inesperadas,assegurando a continuidade da operação. / Salvador

CAPES / O conforto e a liberdade de movimentos de pacientes com doenças crônicas e que têm que ser continuamente monitorados é um tema que tem incentivado o desenvolvimento de novas tecnologias como as redes de sensores corporais sem fios (WBAN) e novas áreas de pesquisa como a telemedicina. Além disso, a incorporação de software inteligente que permite simular o raciocínio dos especialistas, auxiliá-los na tomada de decisões e detectar com antecedência condições anormais ou tendência ao desenvolvimento de determinadas doenças, abre um campo ainda maior de pesquisas, como o campo da Inteligência Artificial na Medicina (AIM). O monitoramento de pacientes por meio de equipamentos sem fios, em conjunto com a tecnologia AIM, permite desenvolver soluções práticas para monitorar pacientes sem descuidar de seu conforto. Nesta tese foram pesquisadas técnicas inteligentes para o desenvolvimento de uma aplicação que permita monitorar cinco sinais vitais de pacientes sem que eles precisem usar leitos hospitalares. Em uma primeira etapa, os procedimentos médicos tipicamente usados pelos especialistas para avaliar um paciente foram estudados e transformados em regras para o modelo fuzzy. O modelo fuzzy proposto permite analisar o estado clínico presente do paciente e criar as saídas desejadas (targets) que permitam treinar as redes neurais artificiais. Posteriormente foi desenvolvido um modelo neural que, analisando os dados atuais e saídas anteriores do paciente, permite prever o seu estado clínico futuro próximo. A fim de achar a metodologia mais exata, cinco redes neurais artificiais foram analisadas e comparadas umas às outras. As redes Elman MISO, Elman MIMO, e NNARX – totalmente conectadas e podadas – foram testadas. O modelo fuzzy teve um excelente resultado concordando com as respostas dadas pelos especialistas em 99,76% dos casos. Depois de analisar as redes propostas no conjunto de validação, os resultados revelaram que unicamente a rede NNARX podada pode oferecer a mais alta acurácia de 99,82%, enquanto os outros modelos degradam o seu desempenho em até 35%. As técnicas de parada antecipada para o treinamento junto com a obtenção de valores médios de MSE, FPE e coeficientes de correlação conseguiram obter as melhores topologias de cada tipo de rede, fazendo quase desnecessária a sua poda. As redes NNARX e P-NNARX conseguiram resultados bem melhores que as redes restantes, mas a acurácia na rede P-NNARX observou um aumento de 1,27% em relação à rede NNARX. Como conclusão, pode-se dizer que, para este caso particular, as redes NNARX capturam a essência do sistema dinâmico não linear muito melhor do que as redes Elman. Finalmente, a rede P-NNARX foi a escolhida para a implementação do sistema inteligente proposto nesta tese. A sua acurácia foi de 99,25% para uma predição no tempo (t + d), onde d = 1 segundo, utilizando os dados de 30 novos pacientes. Foram feitas mais provas com periodos de predição maiores e o sistema demostrou uma ligeira diminuição na acurácia, chegando a 94,58% para d = 60 segundos, mas ainda ficando na faixa dos 90%. Os resultados demonstram o alto nível de generalização do sistema e o excelente desempenho na predição dos três estados clínicos do paciente (estável, semiestável e instável). Pretende-se que este sistema inteligente possa ser usado como ferramenta para a medicina preventiva em pacientes crônicos. / The comfort and freedom of movements of patients that have to be continually monitored is a theme that has motivated the development of new technologies such as networks of wireless body sensors (WBAN) and new research areas such as telemedicine. In addition, the incorporation of intelligent software to simulate the reasoning of experts, assist them in decision making and in early detection of abnormal conditions or tendencies to develop certain diseases, opens an even larger field of research, such as the field of Artificial Intelligence in Medicine (AIM beings its acronym in English). Patient monitoring through wireless equipment and AIM technology allows to develop practical solutions to control patients in environments outside of clinics or hospitals. In this thesis, intelligent tools were used for the development of an application that allows monitoring of five vital signs of patients without them being present in a hospital bed. In a first step, typical medical procedures used by specialists for evaluating a patient were studied and transformed into rules for the fuzzy model. The proposed fuzzy model allows the analysis of the current state of the patient to create the desired outputs (targets) that are used to train the artificial neural networks. Then, a neural model was developed which, by analysing current and historic patient data, forecasts patients’ clinical status in the near future. In order to find the most exact methodology, five artificial neural networks were analyzed and compared with each other using thousands of real patient data sets. Elman MISO, Elman MIMO and NNARX – fully connected and pruned – were tested. The fuzzy model answered in a excelent form, agreeing in 99.76% to the answers given by the experts. After analizing the proposed networks in the validation dataset, it was discovered that the pruned NNARX can offer the highest overall accuracy of 99.82%, whereas the others show a decrease of up to 35%. Through techniques such as early stopping for the training with the search of the mean of MSE, FPE and correlation coefficients it was possible to achieve the best topologies of every network type, making their pruning almost unnecessary. The fully connected NNARX and the P-NNARX achieved much better results than other networks, but an increase of 1.27% was observed in the overall accuracy of the pruned network with respect to the NNARX. It can be said that for this particular case, NNARX networks capture the essence of the non-linear dynamic system much better than Elman. Finally, the P-NNARX model was chosen for the implementation of the proposed smart system. Its overall acuracy was of 99.25%, for the prediction time (t + d), with d = 1 second, by using unseen data of 30 new patients. More tests made with longer prediction periods demonstrate a slight decrease in the overall accuracy reaching up to 94.58% for d = 60 seconds. Nevertheless, it still remained over 90%. Results demonstrate the high generalization level of the system and its excellent performance in predicting the three possible patient conditions (stable, semi-stable, unstable). The next step is to turn this intelligent system into an usefull tool for preventive medicine for chronic patients.

CAPES / O conforto e a liberdade de movimentos de pacientes com doenças crônicas e que têm que ser continuamente monitorados é um tema que tem incentivado o desenvolvimento de novas tecnologias como as redes de sensores corporais sem fios (WBAN) e novas áreas de pesquisa como a telemedicina. Além disso, a incorporação de software inteligente que permite simular o raciocínio dos especialistas, auxiliá-los na tomada de decisões e detectar com antecedência condições anormais ou tendência ao desenvolvimento de determinadas doenças, abre um campo ainda maior de pesquisas, como o campo da Inteligência Artificial na Medicina (AIM). O monitoramento de pacientes por meio de equipamentos sem fios, em conjunto com a tecnologia AIM, permite desenvolver soluções práticas para monitorar pacientes sem descuidar de seu conforto. Nesta tese foram pesquisadas técnicas inteligentes para o desenvolvimento de uma aplicação que permita monitorar cinco sinais vitais de pacientes sem que eles precisem usar leitos hospitalares. Em uma primeira etapa, os procedimentos médicos tipicamente usados pelos especialistas para avaliar um paciente foram estudados e transformados em regras para o modelo fuzzy. O modelo fuzzy proposto permite analisar o estado clínico presente do paciente e criar as saídas desejadas (targets) que permitam treinar as redes neurais artificiais. Posteriormente foi desenvolvido um modelo neural que, analisando os dados atuais e saídas anteriores do paciente, permite prever o seu estado clínico futuro próximo. A fim de achar a metodologia mais exata, cinco redes neurais artificiais foram analisadas e comparadas umas às outras. As redes Elman MISO, Elman MIMO, e NNARX – totalmente conectadas e podadas – foram testadas. O modelo fuzzy teve um excelente resultado concordando com as respostas dadas pelos especialistas em 99,76% dos casos. Depois de analisar as redes propostas no conjunto de validação, os resultados revelaram que unicamente a rede NNARX podada pode oferecer a mais alta acurácia de 99,82%, enquanto os outros modelos degradam o seu desempenho em até 35%. As técnicas de parada antecipada para o treinamento junto com a obtenção de valores médios de MSE, FPE e coeficientes de correlação conseguiram obter as melhores topologias de cada tipo de rede, fazendo quase desnecessária a sua poda. As redes NNARX e P-NNARX conseguiram resultados bem melhores que as redes restantes, mas a acurácia na rede P-NNARX observou um aumento de 1,27% em relação à rede NNARX. Como conclusão, pode-se dizer que, para este caso particular, as redes NNARX capturam a essência do sistema dinâmico não linear muito melhor do que as redes Elman. Finalmente, a rede P-NNARX foi a escolhida para a implementação do sistema inteligente proposto nesta tese. A sua acurácia foi de 99,25% para uma predição no tempo (t + d), onde d = 1 segundo, utilizando os dados de 30 novos pacientes. Foram feitas mais provas com periodos de predição maiores e o sistema demostrou uma ligeira diminuição na acurácia, chegando a 94,58% para d = 60 segundos, mas ainda ficando na faixa dos 90%. Os resultados demonstram o alto nível de generalização do sistema e o excelente desempenho na predição dos três estados clínicos do paciente (estável, semiestável e instável). Pretende-se que este sistema inteligente possa ser usado como ferramenta para a medicina preventiva em pacientes crônicos. / The comfort and freedom of movements of patients that have to be continually monitored is a theme that has motivated the development of new technologies such as networks of wireless body sensors (WBAN) and new research areas such as telemedicine. In addition, the incorporation of intelligent software to simulate the reasoning of experts, assist them in decision making and in early detection of abnormal conditions or tendencies to develop certain diseases, opens an even larger field of research, such as the field of Artificial Intelligence in Medicine (AIM beings its acronym in English). Patient monitoring through wireless equipment and AIM technology allows to develop practical solutions to control patients in environments outside of clinics or hospitals. In this thesis, intelligent tools were used for the development of an application that allows monitoring of five vital signs of patients without them being present in a hospital bed. In a first step, typical medical procedures used by specialists for evaluating a patient were studied and transformed into rules for the fuzzy model. The proposed fuzzy model allows the analysis of the current state of the patient to create the desired outputs (targets) that are used to train the artificial neural networks. Then, a neural model was developed which, by analysing current and historic patient data, forecasts patients’ clinical status in the near future. In order to find the most exact methodology, five artificial neural networks were analyzed and compared with each other using thousands of real patient data sets. Elman MISO, Elman MIMO and NNARX – fully connected and pruned – were tested. The fuzzy model answered in a excelent form, agreeing in 99.76% to the answers given by the experts. After analizing the proposed networks in the validation dataset, it was discovered that the pruned NNARX can offer the highest overall accuracy of 99.82%, whereas the others show a decrease of up to 35%. Through techniques such as early stopping for the training with the search of the mean of MSE, FPE and correlation coefficients it was possible to achieve the best topologies of every network type, making their pruning almost unnecessary. The fully connected NNARX and the P-NNARX achieved much better results than other networks, but an increase of 1.27% was observed in the overall accuracy of the pruned network with respect to the NNARX. It can be said that for this particular case, NNARX networks capture the essence of the non-linear dynamic system much better than Elman. Finally, the P-NNARX model was chosen for the implementation of the proposed smart system. Its overall acuracy was of 99.25%, for the prediction time (t + d), with d = 1 second, by using unseen data of 30 new patients. More tests made with longer prediction periods demonstrate a slight decrease in the overall accuracy reaching up to 94.58% for d = 60 seconds. Nevertheless, it still remained over 90%. Results demonstrate the high generalization level of the system and its excellent performance in predicting the three possible patient conditions (stable, semi-stable, unstable). The next step is to turn this intelligent system into an usefull tool for preventive medicine for chronic patients.