Este trabalho apresenta uma extensão à formulação multivariável do método de controle baseado em dados conhecido como o Método da Referência Virtual, ou Virtual Reference Feedback Tuning (VRFT). Ao lidar com processos onde o ruído é significativo, as formulações tradicionais do VRFT, por mínimos quadrados ou variáveis instrumentais, apresentam propriedades estatísticas insatisfatórias, que acabam levando o sistema de controle em malha fechada a desempenhos muito distantes daqueles especificados pelo projetista. Portanto, visando aprimorar a qualidade destas estimativas e, consequentemente, os desempenhos em malha fechada, esta dissertação propõe a adição de regularização no método VRFT para sistemas multivariáveis. Regularização é uma ferramenta que vem sendo amplamente utilizada e desenvolvida nos últimos anos nas comunidades de Identificação de Sistemas e Machine Learning e é indicada para reduzir a alta covariância que existe nas estimativas - problema que ocorre na formulação do VRFT com variáveis instrumentais. Também, como contribuições deste trabalho destacam-se uma análise mais detalhada do problema de identificação com regularização para sistemas multivariáveis, assim como o desenvolvimento da matriz ótima de regularização para este cenário e as propriedades da nova formulação do VRFT. Para demonstrar a eficiência desta nova formulação do VRFT são desenvolvidos exemplos numéricos. / This work proposes a new extension for the multivariable formulation of the datadriven control method known as Virtual Reference Feedback Tuning. When the process to be controlled contains a significant amount of noise, the standard VRFT approach, that uses either the least squares method or the instrumental variable technique, yield estimates with very poor statistical properties, that may lead the control system to undesirible closed loop performances. Aiming to enhance these statistical properties and hence, the system’s closed loop performance, this work proposes the use of regularization on the multivariable formulation of the VRFT method. Regularization is a feature that has been widely used and researched on the System Identification and Machine Learning communities on the last few years, and it is well suited to cope the high variance issue that emerge on the VRFT method with instrumental variable. Also, a more detailed analysis on the use of regularization for identification of multivariable systems, the proof of the optimal regularization matrix and the exposure of the new regularized VRFT properties can be highlighted as novelties of this work.

Este trabalho trata da utilização de um método de sintonia de controladores baseado nos dados obtidos da planta. A proposta é a sintonia de controladores ressonantes para aplicação em inversores de frequência presentes em fontes ininterruptas de energia, com o intuito de seguimento de referência senoidal de tensão. Dentro deste contexto, será usado o algoritmo Virtual Reference Feedback Tuning, o qual é um método de identificação de controladores baseado em dados que não é iterativo e não necessita do modelo do sistema para identificar o controlador. A partir dos dados obtidos da planta e também da definição de um modelo de referência pelo projetista, o método estima os parâmetros de uma estrutura fixada previamente para o controlador através da minimização de uma função custo definida pelo erro entre a saída desejada e a saída real. Além disso, uma realimentação de corrente é necessária na malha de controle, onde seu ganho proporcional é definido por experimento empírico. Para demonstrar a utilização do método são apresentados resultados simulados e práticos de uma fonte ininterrupta de energia com potência de 5 kV A utilizando cargas lineares e não-lineares. É avaliado o desempenho do ponto de vista da qualidade do sinal de saída real obtido com controladores sintonizados a partir de diferentes modelos de referência, além do uso de sinais de excitação diversos para o algoritmo V RFT. Os resultados experimentais são obtidos em um inversor de frequência monofásico com uma plataforma em tempo real baseada na placa de aquisição de dados dSPACE DS1104. Os resultados mostram que, em relação as normas internacionais, o sistema de controle proposto possui bom comportamento para seguimento de referência, operando à vazio ou utilizando carga linear. / This work discusses about controller tuning methods based on plant data. The proposal is to tune resonant controllers for application to the frequency inverters found in uninterruptible power supplies, with the goal of following sinusoidal reference signals. Within this context, the Virtual Reference Feedback Tuning algorithm is used, which is a data-driven controller identification method that is not iterative and does not require a system model to identify the controller. Data obtained from the plant and also the definition of a reference model by the designer, are used by the method to estimate the parameters of a previously fixed controller structure through the minimization of a cost function, which is defined by the error between desired and actual outputs. Moreover, a current feedback is required in the control loop where the proportional gain is defined by empirical experiment. To demonstrate the method’s application, simulated and practical results of an uninterruptible power supply with capacity of the 5 kV A will be presented employing linear and nonlinear loads. Evaluates the performance in terms of system’s actual output quality, obtained with controllers tuned with different reference models. Distinct excitation signals are also used to feed the VRFT algorithm. The experimental results achieved from use of an single-phase inverter and a real-time platform based on data acquisition board dSPACE DS1104. The results show that, with respect to international standards, the proposed control system has good performance for tracking reference, operating at empty or using linear load.

Este trabalho trata da utilização de um método de sintonia de controladores baseado nos dados obtidos da planta. A proposta é a sintonia de controladores ressonantes para aplicação em inversores de frequência presentes em fontes ininterruptas de energia, com o intuito de seguimento de referência senoidal de tensão. Dentro deste contexto, será usado o algoritmo Virtual Reference Feedback Tuning, o qual é um método de identificação de controladores baseado em dados que não é iterativo e não necessita do modelo do sistema para identificar o controlador. A partir dos dados obtidos da planta e também da definição de um modelo de referência pelo projetista, o método estima os parâmetros de uma estrutura fixada previamente para o controlador através da minimização de uma função custo definida pelo erro entre a saída desejada e a saída real. Além disso, uma realimentação de corrente é necessária na malha de controle, onde seu ganho proporcional é definido por experimento empírico. Para demonstrar a utilização do método são apresentados resultados simulados e práticos de uma fonte ininterrupta de energia com potência de 5 kV A utilizando cargas lineares e não-lineares. É avaliado o desempenho do ponto de vista da qualidade do sinal de saída real obtido com controladores sintonizados a partir de diferentes modelos de referência, além do uso de sinais de excitação diversos para o algoritmo V RFT. Os resultados experimentais são obtidos em um inversor de frequência monofásico com uma plataforma em tempo real baseada na placa de aquisição de dados dSPACE DS1104. Os resultados mostram que, em relação as normas internacionais, o sistema de controle proposto possui bom comportamento para seguimento de referência, operando à vazio ou utilizando carga linear. / This work discusses about controller tuning methods based on plant data. The proposal is to tune resonant controllers for application to the frequency inverters found in uninterruptible power supplies, with the goal of following sinusoidal reference signals. Within this context, the Virtual Reference Feedback Tuning algorithm is used, which is a data-driven controller identification method that is not iterative and does not require a system model to identify the controller. Data obtained from the plant and also the definition of a reference model by the designer, are used by the method to estimate the parameters of a previously fixed controller structure through the minimization of a cost function, which is defined by the error between desired and actual outputs. Moreover, a current feedback is required in the control loop where the proportional gain is defined by empirical experiment. To demonstrate the method’s application, simulated and practical results of an uninterruptible power supply with capacity of the 5 kV A will be presented employing linear and nonlinear loads. Evaluates the performance in terms of system’s actual output quality, obtained with controllers tuned with different reference models. Distinct excitation signals are also used to feed the VRFT algorithm. The experimental results achieved from use of an single-phase inverter and a real-time platform based on data acquisition board dSPACE DS1104. The results show that, with respect to international standards, the proposed control system has good performance for tracking reference, operating at empty or using linear load.

Este trabalho trata da utilização de um método de sintonia de controladores baseado nos dados obtidos da planta. A proposta é a sintonia de controladores ressonantes para aplicação em inversores de frequência presentes em fontes ininterruptas de energia, com o intuito de seguimento de referência senoidal de tensão. Dentro deste contexto, será usado o algoritmo Virtual Reference Feedback Tuning, o qual é um método de identificação de controladores baseado em dados que não é iterativo e não necessita do modelo do sistema para identificar o controlador. A partir dos dados obtidos da planta e também da definição de um modelo de referência pelo projetista, o método estima os parâmetros de uma estrutura fixada previamente para o controlador através da minimização de uma função custo definida pelo erro entre a saída desejada e a saída real. Além disso, uma realimentação de corrente é necessária na malha de controle, onde seu ganho proporcional é definido por experimento empírico. Para demonstrar a utilização do método são apresentados resultados simulados e práticos de uma fonte ininterrupta de energia com potência de 5 kV A utilizando cargas lineares e não-lineares. É avaliado o desempenho do ponto de vista da qualidade do sinal de saída real obtido com controladores sintonizados a partir de diferentes modelos de referência, além do uso de sinais de excitação diversos para o algoritmo V RFT. Os resultados experimentais são obtidos em um inversor de frequência monofásico com uma plataforma em tempo real baseada na placa de aquisição de dados dSPACE DS1104. Os resultados mostram que, em relação as normas internacionais, o sistema de controle proposto possui bom comportamento para seguimento de referência, operando à vazio ou utilizando carga linear. / This work discusses about controller tuning methods based on plant data. The proposal is to tune resonant controllers for application to the frequency inverters found in uninterruptible power supplies, with the goal of following sinusoidal reference signals. Within this context, the Virtual Reference Feedback Tuning algorithm is used, which is a data-driven controller identification method that is not iterative and does not require a system model to identify the controller. Data obtained from the plant and also the definition of a reference model by the designer, are used by the method to estimate the parameters of a previously fixed controller structure through the minimization of a cost function, which is defined by the error between desired and actual outputs. Moreover, a current feedback is required in the control loop where the proportional gain is defined by empirical experiment. To demonstrate the method’s application, simulated and practical results of an uninterruptible power supply with capacity of the 5 kV A will be presented employing linear and nonlinear loads. Evaluates the performance in terms of system’s actual output quality, obtained with controllers tuned with different reference models. Distinct excitation signals are also used to feed the VRFT algorithm. The experimental results achieved from use of an single-phase inverter and a real-time platform based on data acquisition board dSPACE DS1104. The results show that, with respect to international standards, the proposed control system has good performance for tracking reference, operating at empty or using linear load.

A software component model is the cornerstone of any Component-based Software Development (CBSD) methodology. Such a model defines the modelling elements for constructing software systems. In software system modelling, it is necessary to capture the three elements of a system's behaviour: (i) control (ii) computation and (iii) data. Within a system, computations are performed according to the flow of control or the flow of data, depending on whether computations are control-driven or data-driven. Computations are function evaluations, assignments, etc., which transform data when invoked by control or data flow. Therefore a component model should be able to model control flow, data flow as well as computations. Current component models all model computations, but beside computations tend to model either control flow only or data flow only, but not both. In this thesis, we present a new component model which can model both control flow and data flow. It contains modelling elements that capture control flow and data flow explicitly. Furthermore, the modelling of control flow is separate from that of data flow; this enables the modelling of both control-driven and data-driven computations. The feasibility of the model is shown by means of an implementation of the model, in the form of a prototype tool. The usefulness of the model is then demonstrated for a specific domain, the embedded systems domain, as well as a generic domain. For the embedded systems domain, unlike current models, our model can be used to construct systems that are both control-driven and data-driven. In a generic domain, our model can be used to construct domain models, by constructing control flows and data flows which together define a domain model.

With the recent advances in the field of unmanned aerial vehicles, many applications have been identified. In tasks that require high-payload-to-weight ratios, flight times in the order of days, reduced noise and/or hovering capabilities, lighter-than-air vehicles present themselves as a competitive platform compared to fixed-wing and rotor based vehicles. The limiting factor in their widespread use in autonomous applications comes from the complexity of the control task. The so-called airships are highly-susceptible to aerodynamic forces and pose complex nonlinear system dynamics that complicate their modeling and control. Model-free control lends itself well as a solution to this type of problem, as it derives its control policies using input-output data, and can therefore learn complex dynamics and handle uncertain or unknown parameters and disturbances. In this work, two multi-input multi-output algorithms are presented on the basis of optimal control theory. Leveraging results from reinforcement learning, a single layer, partially connected neural network is formulated as a value function appropriator in accordance with Weierstrass higher-order approximation theorem. The so-called critic-network is updated using gradient descent methods on the mean-squared error of the temporal difference equation. In the single-network controller, the control policy is formulated as a closed form equation that is parameterized on the weights of the critic-network. A second controller is proposed that uses a second single-layer partially connected neural network, the actor-network, to calculate the control action. The actor-network is also updated using gradient descent on the squared error of the temporal difference equation. The controllers are employed in a highly realistic simulation airship model in nominal conditions and in the presence of external disturbances in the form of turbulent wind. To verify the validity and test the sensitivity of the algorithms to design parameters (the initialization of certain terms), ablation studies are carried out with multiple initial parameters. Both of the proposed algorithms are able to track the desired waypoints in both the nominal and disturbed flight tests. Furthermore, the performance of the controllers is compared to a modern, state-of-the-art multi-input multi-output controller. The two proposed controllers outperform the comparison controller in all but one flight test, with up to four fold reduction in the integral absolute error and integral time absolute error metrics. On top of the quantitative improvements seen in the proposed controllers, both controllers demonstrate a reduction in system oscillation and actuator chattering with respect to the comparison algorithm.

Smart cities embrace cutting-edge technologies to improve transportation efficiency and safety. With the rollout of 5G and an ever-growing network of connected infrastructure sensors, real-time road condition awareness is becoming a reality. However, this progress brings new challenges. The communication and vast amounts of data generated by autonomous vehicles and the connected infrastructure must be navigated. Furthermore, different levels of autonomous driving (ranging from 0 to 5) are rolled out gradually and human-driven vehicles will continue to share the roads with autonomous vehicles for some time. In this work, we apply a data-driven control scheme called Learning Model Predictive Control (LMPC) to three different smart city scenarios of increasing complexity. Given a successful execution of a scenario, LMPC uses the trajectory data from previous executions to improve the performance of subsequent executions while guaranteeing safety and recursive feasibility. Furthermore, the performance from one execution to another is guaranteed to be non-decreasing. For our three smart-city scenarios, we apply a minimum time objective and start with a single vehicle in a two-lane intersection. Then, we add an obstacle on the lane of the ego vehicle, and lastly, we add oncoming traffic. We find that LMPC gives us improved traffic efficiency with shorter travel. However, we find that LMPC may not be suitable for real-time training in smart city scenarios. Thus, we conclude that this approach is suitable for simulator-driven, offline, training on any trajectory data that might be generated from autonomous vehicles and the infrastructure sensors in future smart cities. Over time, this can be used to construct large data sets of optimal trajectories which are available for the connected vehicles in most urban scenarios. / Smarta städer använder modern teknik för att förbättra transporteffektiviteten och säkerheten. Med införandet av 5G och ett allt större nätverk av uppkopplade sensorsystem för infrastruktur blir realtidsmedvetenhet om vägförhållandena en verklighet. Denna utveckling medför nya utmaningar. Kommunikationen mellan autonoma fordon och uppkopplade sensorsystem ger upphov till stora mängder data som måste hanteras. Dessutom kommer fordon med olika autocnominivåer (från 0 till 5) att behöva dela gatorna tillsammans med människostyrda fordon samtidigt under en tid. I detta arbete tillämpar vi en datadriven reglermetod som heter Learning Model Predictive Control (LMPC) på tre olika scenarier i en smart stad med ökande komplexitet. LMPC utnyttjar data från en tidigare lyckad körning av ett visst scenario för att förbättra prestandan på efterföljande körningar samtidigt som säkerheten och rekursiv genomförbarhet garanteras. Vidare garanteras att prestandan från en körning till en annan inte minskar. För våra tre scenarier är målet att minimerar restiden och börjar med ett enda fordon i en tvåfilig korsning. Sedan lägger vi till ett hinder på högra filen och till sist lägger vi till mötande trafik. Vi finner att LMPC ger oss förbättrad trafikeffektivitet med kortare restid. Vi finner dock att LMPC må vara mindre lämplig för realtids scenarier. Således drar vi slutsatsen att denna metod är lämplig för optimering i simulatorer, offline, på data som kan genereras från autonoma fordon och sensorsystemet i infrastrukturen. Så småningom kan vår metod användas för att konstruera stora dataset av optimala trajektorier som är tillgängliga för uppkopplade fordon i framtidens smarta städer.

With the development of technology and availability of data, it is sometimes easier to learn the control policies directly from the data, rather than modeling a plant and designing a controller. Modeling a plant is not always possible due to the complexity of the plant. Data-enabled predictive control (DeePC) is a recently proposed approach that combines system identification, estimation, and control in a single optimization problem. DeePC is primarily designed for LTI systems. The purpose of this thesis is to extend the application of DeePC to nonlinear systems with a particular focus on a non-holonomic ground robot. To reach this goal, we decompose the system states into different working modes where each mode can be linearly approximated. Furthermore, the data collection policies were also evaluated to conclude how they affect the performance of the DeePC. We identified several key challenges in this direction, namely: data-demanding structure, high computational complexity, and performance deterioration with increased non-linearity. While these challenges prohibited the application of DeePC to the ground robot system; we successfully applied the method to a benchmark non-linear system, the inverted pendulum on cart problem, and studied the effect of various design choices on control performance. Our observations indicate potential areas of improvement toward enabling DeePC for highly nonlinear systems. / Med utvecklingen av teknik och tillgänglighet av data är det ibland enklare att lära sig styrpolicyerna direkt från data, snarare än att modellera ett system och designa en styrenhet. Att modellera ett system är inte alltid möjligt på grund av systemets komplexitet. Data aktiverad prediktiv kontroll (DeePC) är en nyligen föreslagen metod som kombinerar systemidentifiering, uppskattning och kontroll i ett enda optimeringsproblem. DeePC är främst designad för LTI-system. Syftet med denna avhandling är att utöka tillämpningen av DeePC till icke-linjära system med särskilt fokus på en icke-holonomisk markrobot. För att nå detta mål delar vi upp systemtillstånden i olika arbetslägen där varje läge kan approximeras linjärt. Dessutom utvärderades datainsamlingspolicyerna för att dra slutsatser om hur de påverkar DeePCs prestation. Vi identifierade ett antal nyckelutmaningar i denna riktning, nämligen: datakrävande struktur, hög beräkningskomplexitet och prestandaförsämring med ökad icke-linjäritet. Även om de utmaningerna hindrade tillämpningen av DeePC på markrobot systemet; har vi framgångsrikt tillämpat metoden på ett benchmark icke-linjärt system, problemet med inverterad pendel på vagn, och studerade effekten av olika designval på kontrollprestanda. Våra observationer indikerar potentiella förbättringsområden för att möjliggöra DeePC för mycket olinjära system.