Fundação Araucária / Este trabalho apresenta o estudo, projeto e simulação de um conversor CC boost, com controle da tensão de saída utilizando redes neurais. O conversor boost estudado neste trabalho está sendo utilizado em um aerogerador de baixa potência, com potência máxima de 3kW. Devido à operação em velocidade variável do aerogerador, a tensão de entrada no conversor também é variável em uma ampla faixa de operação, e a saída deve ser estabilizada para uma tensão fixa. Isso faz com que haja a necessidade do controle do conversor para que as diferenças na tensão de entrada sejam compensadas. Para a compensação dessa diferença, foi projetado um controlador utilizando o método de lugar das raízes. A modelagem do conversor boost também é apresentada. O controlador fornece o tempo de atuação da chave semicondutora de potência utilizada no conversor, controlando assim a tensão de saída. A variação na tensão de entrada, que pode variar de 50V a 350V, faz com que o controlador não atue de forma otimizada para todos os pontos de operação. Dessa forma, um controlador neural foi projetado para que trabalhe como controlador, compensando distúrbios de tensão de entrada do conversor. A validação dos controles foi efetuada através de simulação utilizando o software Matlab/Simulink, para confirmação do desenvolvimento teórico apresentado no trabalho. / This work presents the study, design and simulation of a DC boost converter with output voltage control using neural networks. The boost converter studied in this work is being used in a wind turbine low power, with maximum power of 3kW. Due to the operation of the variable speed of the wind turbine, the input voltage of the converter is also variable over a wide operating range, and the output must be stabilized to a fixed voltage. This means that there is a need to control the converter so that differences in input voltage are compensated. To compensation this difference, a controller was designed using the root locus method. The modeling of the boost converter is also presented. The controller provides the time of operation of power semiconductor switch used in the converter, thereby controlling the output voltage. The variation in input voltage, which can vary from 50V to 350V, makes the driver does not act optimally for all operating points. Thus, a neural controller is designed to work as a controller, compensating disturbances of the converter input voltage. The validation of this controller was performed by simulation using Matlab / Simulink software, to confirm the theoretical development presented in the work.

Electrical Power System is one of the most complex real time operating systems. It is probably one of the best examples of a large interconnected nonlinear system of varying nature. The system needs to be operated and controlled with component or system problems, often with combinatorial complexity. In addition, time scales of operation and control can vary from milliseconds to minutes to hours. It is difficult to maintain such a system at constant operating condition due to both small and large disturbances such as sudden change in loads, change in network configuration, fluctuations in turbine output, and various types of faults etc. The system is therefore affected by a variety of instability problems. Among all these instability problems one of the important modes of instability is related to dynamic instability or more precisely the small perturbation oscillatory instability. Oscillations of small magnitude and low frequency (in the range of 0.1Hz to 2.5Hz) could persist for long periods, limiting the power transfer capability of the transmission lines. Power System Stabilizers (PSS) were developed as auxiliary controllers on the excitation system to improve the system damping performance by modulating the generator excitation voltage. However, the synthesis of an effective PSS for all operating conditions still remains a difficult and challenging task. The design and tuning of PSS for robust operation is a laborious process. The existing PSS design techniques require considerable expertise, the complete system information and extensive eigenvalue calculations which increases the computational burden as the system size increases. Conventional automatic voltage regulator (AVR) and PSS designs are based on linearized models of power systems which fail to stabilize the system over a wide range of operating conditions. In the last decade or so, a variety of nonlinear control techniques have become available. In this thesis, an attempt is made to explore the suitability of some of these design techniques for designing excitation controllers to enhance small perturbation stability of power systems over a wide range of operating and system conditions. This thesis first proposes a method of designing power system stabilizers based on local measurements alone, in multi-machine systems. Next, a method has been developed to analyze and quantify the small signal performance benefits of replacing the existing AVR+PSS structure with nonlinear voltage regulators. A number of new nonlinear controller designs have been proposed subsequently. These include, (a) a new decentralized nonlinear voltage regulator for multi machine power systems with a single tunable parameter that can achieve effective trade of between both the voltage regulation and small signal objectives, (b) a decentralized Interconnection and Damping Assignment Passivity Based Controller in addition to a proportional controller that can achieve all the requirements of an excitation system and (c) a Nonlinear Quadratic Regulator PSS using Single Network Adaptive Critic architecture in the frame work of approximate dynamic programming. Performance of all the proposed controllers has been analyzed using a number of multi machine test systems over a range of operating conditions.

In advanced process control, the Model Predictive Control (MPC) may be considered the most important innovation in recent years and the standard tool for industrial applications due to the fact that it keeps the plant operating in the constraints more profitable. However, like every control algorithm, the MPC after some time in operation rarely works as originally designed. Thus, to preserve the benefits of MPC systems for a long period of time, their performance needs to be monitored and evaluated during the operation. This task require the presence of reliable and effective tools to detect when the controller performance is below of the desirable, to define the need, or not, of recommissioning the system. Thus, the objective of this work is development of techniques for monitoring and evaluating the performance of multivariable predictive controllers, being developed two new tools: LQG benchmark Modified and IHMC benchmark. The results obtained from numerical simulations were satisfactory and consistent with the technical literature applied in the developments of the evaluators, which were used in the monitoring of the control system MPC of the oil-water-gas three-phase separation process, offering an appropriate solution and providing subsidies for implementations in real industrial systems. / Em controle avançado de processos, o controlador preditivo ou MPC (Model Predictive Control) pode ser considerado como a mais importante inovação dos últimos anos e a ferramenta padrão para aplicações industriais, devido ao fato do MPC manter a planta operando dentro das suas restrições de forma mais lucrativa. Entretanto, como todo algoritmo de controle, o MPC depois de algum tempo em operação dificilmente funciona como quando fora inicialmente projetado. Desta forma, com o objetivo de manter os benefícios dos sistemas MPC por um longo período de tempo, seu desempenho precisa ser monitorado e avaliado durante a operação. Esta tarefa requer a presença de ferramentas efetivas e confiáveis para detectar quando o desempenho do controlador estiver abaixo do desejável, para definir a necessidade, ou não, de um recomissionamento do sistema. Destarte, aborda-se neste trabalho o desenvolvimento de técnicas para monitoramento e avaliação de desempenho de controladores preditivos multivariáveis, sendo desenvolvidas duas novas ferramentas: LQG benchmark Modificado e IHMC benchmark. Os resultados obtidos a partir de simulações numéricas foram satisfatórios e coerentes com a literatura técnica aplicada no desenvolvimento dos avaliadores, os quais foram utilizados no monitoramento do sistema de controle MPC do processo de separação trifásica água-óleo-gás, oferecendo assim uma solução apropriada e fornecendo subsídios para implementações em sistemas industrias reais.