[pt] Os avanços da Robótica nas últimas décadas permitem um aumento nas gamas de aplicações de manipuladores robóticos em diversos setores da indústria. Isto, impacta diretamente a interação Homem-Robô (HRI), resultando em um aumento de tarefas que requerem compartilhamento de ambiente de trabalho, desempenho de segurança e a habilidade de detecção de contato do manipulador robótico. Consequentemente, métodos de controle capazes de prever contato, controlar força ou trajetória para evitar danos durante colisões se tornam cada vez mais necessários seja por questões de segurança ou de desempenho. Separando a dinâmica de um manipulador de um único elo em dois modos, sendo eles modo de controle de posição (modo livre) e modo de controle de torque (modo de contato), a primeira parte desta dissertação, lida com o problema de estimação de estados para detecção do modo ativo através da implementação do método de Estimação de Estados de Horizonte móvel com Redes Neurais (NNMHSE). A efetividade do método de estimação proposto é avaliada através da comparação dos estados e modos gerados pelo MHSE e dos estimados pela Rede Neural. Este método apresentou baixos valores de RMSE, altos valores de R(2), e uma redução do tempo de processamento do algoritmo de estimação. A segunda parte desta dissertação lida com o problema de controle de posição e força chaveado para um manipulador robótico não linear, aplicando Controle Preditivo Baseado em Modelo (MPC). O algoritmo MPC chaveado implementado mostrou-se capaz de controlar efetivamente ambos os modos do sistema apresentando baixo erro na predição, aproximadamente 2 por cento no modo de controle de posição e 0.5 por cento no modo de controle de torque, mesmo considerando alterações cíclicas nos modos. Ambos os métodos provam ser adequados para controle de manipuladores robóticos colocalizados com seres humanos ou em ambientes desestruturados por meio da detecção do modo de operação e do controle chaveado posição-torque. / [en] The advances in Robotics in recent decades allow a growing range of robotic manipulator applications in various industry sectors. This directly impacts Human-Robot Interaction (HRI), increasing tasks that require a shared work environment, safety performance, and the contact detection ability of the robotic manipulator. Consequently, control methods capable of predicting contact, and controlling force or trajectory to avoid damage during collisions become increasingly necessary either for safety or performance reasons. Separating the dynamics of a single-link manipulator into two modes, namely position control mode (free mode) and torque control mode (contact mode), the first part of this dissertation deals with the estimation problem of states for active mode detection through the implementation of the Moving Horizon State Estimation with Neural Networks (NNMHSE) method. The effectiveness of the proposed estimation method is evaluated by comparing the states and modes generated by the MHSE and those estimated by the Neural Network. This method showed low RMSE values, high values of R(2), and a reduction in the processing time of the estimation algorithm. The second part of this dissertation deals with the position and force switching problem for a non-linear robotic manipulator, applying Model-Based Predictive Control (MPC). The implemented switched MPC algorithm effectively controlled both modes of the system, presenting low prediction error, approximately 2 percent in position control mode and 0.5 percent in torque control mode, even considering cyclical changes in the modes. Both methods prove to be suitable for controlling co-located robotic manipulators with humans or in unstructured environments through operation mode detection and position-torque switching control.

[pt] Os veículos automatizados devem trafegar em determinado ambiente detectando, planejando e seguindo uma trajetória segura. De modo a se mostrarem mais seguros que seres humanos, eles devem ser capazes de executar essas tarefas tão bem ou melhor do que motoristas humanos sob diferentes condições críticas. Uma parte essencial no estudo de veículos automatizados o desenvolvimento de modelos representativos que sejam precisos e computacionalmente eficientes. Assim, para lidar com esses problemas, o presente trabalho aplica métodos de inteligência computational e identificação de sistemas para realizar modelagem de veículos e controle de rastreamento de trajetória. Primeiro, arquiteturas neurais são usadas para capturar as características do pneu na interação entre a dinâmica lateral e longitudinal do veículo, reduzindo o custo computacional em controladores preditivos. Em segundo lugar, uma combinação de modelos caixa-preta é usada para melhorar o controle preditivo. Em seguida, uma abordagem híbrida combina modelos baseados na física e orientados por dados com modelagem de caixa-preta das discrepâncias. Essa abordagem é escolhida para melhorar a precisão da modelagem de veículos, propondo um modelo de discrepância para capturar incompatibilidades entre modelos de veículos e dados medidos. Os resultados são mostrados quando os métodos propostos são aplicados a sistemas com dados simulados/reais e comparados com abordagens encontradas na literatura, mostrando um aumento de precisão (até 40 por cento) em termos de métricas baseadas em erro, com menor esforço computacional (redução de até 88 por cento) do que os controladores preditivos convencionais. / [en] Automated vehicles must travel in a given environment detecting, planning, and following a safe path. In order to be safer than humans, they must be able to perform these tasks as well or better than human drivers under different critical conditions. An essential part of the study of automated vehicles is the development of representative models that are accurate and computationally efficient. Thus, to cope with these problems, the present work applies artificial neural networks and system identification methods to perform vehicle modeling and trajectory tracking control. First, neural architectures are used to capture tire characteristics present in the interaction between lateral and longitudinal vehicle dynamics, reducing computational costs for predictive controllers. Secondly, a combination of black-box models is used to improve predictive control. Then, a hybrid approach combines physics-based and data-driven models with black-box modeling of the discrepancies. This approach is chosen to improve the accuracy of vehicle modeling by proposing a discrepancy model to capture mismatches between vehicle models and measured data. Results are shown when the proposed methods are applied to systems with simulated/real data and compared with approaches found in the literature, showing an increase of accuracy (up to 40 percent) in terms of error-based metrics while having lesser computational effort (reduction by up to 88 percent) than conventional predictive controllers.

[pt] Este trabalho tem como objetivo estudar estratégias de Controle Preditivo baseado em Modelo Não-Linear (NMPC) aplicadas a uma coluna de destilação despropanizadora simulada. Essas colunas são empregadas em unidades de processamento de gás natural (UPGNs) para a separação do produto propano do butano. Colunas de destilação possuem características particularmente desafiadoras sob o ponto de vista de controle, como: não-linearidades, grandes constantes de tempo, atraso, restrições de variáveis e inversão do sinal de ganho estático. Como as medidas de composição frequentemente possuem atrasos e dados esparsos, os sistemas de controle convencionais não são capazes de controlar a composição diretamente e possuem dificuldade em manter os produtos dentro das especificações. Contudo, controladores baseados em modelo possuem a habilidade de prever a composição através do modelo interno do processo, além de serem capazes de lidar com restrições. Na literatura, nenhuma aplicação do modelo de Hammerstein modificado para coluna de destilação ou para sistemas multivariáveis foi encontrada, sendo esta uma novidade. Desta forma, foram estudadas três estratégias de controle: controle PID tradicional, NMPC com modelo de Hammerstein modificado (H-NMPC) e NMPC com modelo por Redes Neurais (NN-NMPC). O sistema estudado foi identificado de forma a se obter valores numéricos adequados aos parâmetros dos modelos. A identificação dos parâmetros dos modelos e os algoritmos de NMPC foram implementados no ambiente MATLAB. A coluna de destilação foi simulada usando o Aspen Plus Dynamics. Como resultado, o H-NMPC teve o melhor desempenho de controle ao rastrear diferentes trajetórias de referência, a desacoplar as variáveis controladas e a rejeitar os distúrbios. Além disso, esta apresentou maior rapidez computacional comparado com a estratégia NNNMPC. / [en] This work aims to study strategies of Nonlinear Model Predictive Control (NMPC) applied to a simulated depropanizer distillation column. These columns are used in natural gas processing units (NGPUs) for the separation of the product propane from butane. Distillation columns have particularly challenging features from the control point of view, such as: nonlinearities, large time constants, delay, variable constraints and static gain signal inversion. Because compositional measures often have delays and sparse data, conventional control systems are not able to control composition directly and have difficulty keeping products within specifications. However, model-based controllers predict composition through the internal process model, besides being able to handle constraints. In the literature, no applications of the modified Hammerstein model for distillation column or multivariable systems was found, so this is a novelty. Therefore, three control strategies were studied: traditional PID control, NMPC with modified Hammerstein model (H-NMPC) and NMPC with neural network model (NN-NMPC). The studied system was identified in order to obtain adequate numerical values of the model parameters. The model identification and the NMPC algorithms were implemented in the MATLAB environment. The distillation column was simulated using Aspen Plus Dynamics. As a result, the H-NMPC provided better control performance for different setpoint tracking, control variables decoupling, and disturbance rejection. Furthermore, it presented faster computational speed compared to NN-NMPC.

Autonomous driving is one of the megatrends in today’s automotive industry. Passengers are expected to do more non-driving tasks in an autonomous driving vehicle. Therefore, the comfort of the vehicle has become a more important factor for the passengers. This thesis investigates the possibility of increasing comfort through an integrated active chassis control strategy. First, this thesis has defined comfort in objective ways. Then, the objective comfort evaluation variables are used for comfort evaluation of the vehicle in different scenarios. The improvement in comfort is evaluated for four active chassis systems, including active suspension, active anti-roll bar, active rear-wheel steering and torque vectoring systems. Since more than one active chassis system can affect vehicle body motion in one direction, those four active chassis systems should be controlled in an integrated way. The model predictive control (MPC) is used because it can control a multi-input multi-output system in an optimized way. Two MPC controllers have been developed in this thesis to control multiple active chassis systems for comfort improvement. The original MPC controller is a linear MPC controller that uses a time-invariant state-space vehicle model. The adaptive MPC controller is a linear MPC controller that uses a time-variant state-space vehicle model. These two controllers are tested in the simulation software CarMaker with various scenarios, such as slalom, double lane-change, and bumps that are both symmetrical and shifted unsymmetrical. Finally, the simulation results are evaluated with objective comfort evaluation methods to assess the controller performances in comfort improvement. In conclusion, the model predictive control can be a feasible way to improve comfort with multiple active chassis systems. The simulation results show that the two MPC controllers can reduce the objective comfort evaluation variables. The discussions of the design process and simulation results point out future works that need to be done before this project becomes a product of real vehicles. / Autonom körning är en av megatrenderna i dagens bilindustri. Passagerare förväntas utföra fler icke-körrelaterade uppgifter i ett autonomt fordon. Därför har fordonets komfort blivit en allt viktigare faktor för passagerarna. Denna avhandling undersöker möjligheten att öka komforten genom en integrerad aktiv chassikontrollstrategi. Som utgångspunkt har denna avhandling definierat komfort på objektiva sätt. Sedan används de objektiva komfortvärderingsvariablerna för komfortutvärdering av fordonet i olika scenarier. Förbättringen av komfort utvärderas för fyra aktiva chassisystem, inkluderande aktiv fjädring, aktiv krängningshämmare, aktiv bakhjulsstyrning och drivmomentvektorisering. Eftersom mer än ett aktivt chassisystem kan påverka fordonets rörelse i en riktning, bör dessa fyra aktiva chassisystem styras på ett integrerat sätt. Modellprediktiv reglering (MPC) används eftersom den kan styra ett multi-input multi-output system på ett optimerat sätt. Två MPC-reglersystem har utvecklats för att styra flera aktiva chassisystem för komfortförbättring. Den ursprungliga MPC-reglerenheten är en linjär MPC-regulator som använder en tidsinvariant fordonsmodell. Den adaptiva MPC-reglerenheten är en linjär MPC-regulator som använder en tidsvariant fordonsmodell. Dessa två reglersystem testas i simuleringsprogramvaran CarMaker i olika scenarier, till exempel slalom, dubbelt körfältsbyte och väg-gupp som är både symmetriska och osymmetriska. Slutligen utvärderas simuleringsresultaten med objektiva komfortutvärderingsmetoder för att bedöma reglersystemens komfortförbättring. Sammanfattningsvis kan modellprediktiv reglering vara ett genomförbart sätt att förbättra komforten med flera aktiva chassisystem. Simuleringsresultaten visar att de två MPC-regulatorerna kan reducera de objektiva komfortutvärderingsvariablerna. Diskussionerna om designprocessen och simuleringsresultaten tar upp framtida arbeten som behöver göras innan detta projekt kan förverkligas i riktiga fordon.

The aim of the thesis was to implement and compare control methods in a model of a main battle tank. Three controllers were implemented in a two axis gimbal model and their performances were compared. The comparisons were performed using step changes in the reference signal, frequency analysis of an oscillating reference signal and disturbances, and turret mass uncertainties. The results showed that the sliding mode controller had the best performance for both reference changes and disturbance attenuation. The PID controller had a better performance for the change in reference, compared to the model predictive controller, but a significantly worse disturbance attenuation. Due to model approximations, such as assuming ideal engines and noise reduction, the results likely show a better performance than what can be expected if applied on a real main battle tank. Therefore, the results show an upper limit of the stabilization performance of turret and barrel control and should only be used to compare the controllers. / Målet med uppsatsen var att implementera och jämföra reglermetoder för en teoretisk modell av en modern stridsvagn. Tre metoder implementerades i ett tvåaxligt gimbalsystem och deras prestanda utvärderades. Mer specifikt utvärderades regulatorernas respons för en referensändring, en referensstörning och osäkerheter i tornets massa och massfördelning. Utifrån dessa resultat jämfördes sedan reglermetoderna. Resultaten visade att sliding mode regulatorn hade bäst prestanda för både referensändring och referensstörningen när man tittar på frekvensanalys av mät- och processstörningar. PID-regulatorn hade en bättre prestanda än MPC-regulatorn för en referensändring men en sämre respons för en referensstö-rning. På grund av modellförenklingar som till exempel antaganden om ideala motorer och brusreduktion visar resultaten troligen en något bättre prestanda än vad som kan uppnås på en riktigt stridsvagn. Det medför att resultaten visar en övre gräns för vad styrsystemet för en stridsvagns torn- och eldrörsrotation kan uppnå.

Cooperative control of autonomous multi-agent systems is a research areawhich is getting significant attention in recent years. Multi-agent systemsallow for a broad spectrum of applications and cooperation can increasetheir flexibility, efficiency and robustness to changes in external constraintsand disturbances. Focusing on autonomous vehicles, examples of possibleapplications of cooperative multi-agent systems include search and rescuemissions, autonomous delivery and performing of emergency landings.The purpose of the thesis is to develop and implement a cooperativerendezvous algorithm based on model predictive control and apply it to theproblem of autonomous landing in an indoor setting. The agents involved in themaneuver are a quadcopter and a ground carrier. The two agents cooperativelynegotiate on the optimal location for the touchdown taking also into accountrelevant spatial constraints and, if necessary, update that location, also referredto as rendezvous point, in real-time throughout the maneuver.The algorithm is first tested and validated in a simulated environment andfinally on the physical system during real-time operations.Additional scenarios are tested in the simulated environment in order tofurther inspect the potential capabilities of the developed algorithm. Thoseadditional simulations analyse how the algorithm behaves when a constantlateral wind influences the quadcopter; when the controllers operate at areduced frequency; and when the quadcopter is affected by an external Gaussiandisturbance.The developed algorithm proved to be suitable for the purpose, allowingthe agents to perform the desired maneuver in a relatively short time and witha high degree of precision. / Kooperativ reglering av autonoma fleragentsystem är ett forskningsområdesom har fått stor uppmärksamhet de senaste åren. Fleragentsystem möjliggörett brett spektrum av applikationer samtidigt som kooperation kan öka derasflexibilitet, effektivitet och robusthet mot förändringar i yttre begränsningar ochstörningar. Med fokus på autonoma fordon, exempel på möjliga tillämpningarav kooperativa fleragentsystem inkluderar sök- och räddningsuppdrag, autonomleverans och utförande av nödlandningar.Syftet med rapporten är att utveckla och implementera en kooperativrendezvous -algoritm baserad på modellprediktiv reglerteknik samt att tillämpaden för att utföra en inomhus autonom landning. I vår uppställning beståragenterna i manövern av en quadcopter och en markbärare. De två agenternaförhandlar samarbetsvilligt om den optimala platsen för landning samtidigtsom de beaktar relevanta rumsliga begränsningar och uppdaterar vid behovden platsen i realtid under hela manövern.Algoritmen testas och valideras först i en simulerad miljö och slutligen pådet fysiska systemet under en realtidsmiljö.Ytterligare scenarier testas i den simulerade miljön för att bortre inspekterapotentialen hos den utvecklade algoritmen. Dessa extra simuleringar illustrerarhur algoritmen beter sig när en konstant sidovind påverkar quadcoptern; närstyrenheterna arbetar med reducerad frekvens; och när quadcoptern påverkasav en yttre Gaussisk störning.Den utvecklade algoritmen visade sig vara lämplig för ändamålet, vilketgjorde att agenterna kunde utföra önskad manöver på relativt kort tid och medhög precision.

Fuel Cell Hybrid Electric vehicles are hybrid vehicles that consist of both fuel cells and batteries as energy conversion systems. The Energy Management System plays an important role in the operation of the fuel cell hybrid system, as it helps in reducing the hydrogen consumption of the system. This study investigates an optimal control algorithm with an aim to reduce the hydrogen consumption of the fuel cell system for five different drive cycles operating in Europe. Model Predictive Control(MPC) is used to solve the optimal control problem, by formalizing a look ahead controller, utilizing its receding horizon approach. The optimal controller analysis is compared with a conventional rule-based controller, by analysing the hybrid system over various battery and fuel cell sizes, on the basis of the overall hydrogen consumption. Firstly, a simplified system model is developed, by modelling the fuel cell system with respect to the efficiency curve of the hydrogen power and fuel cell power. The battery system model with its State of Charge(SOC) is coupled with the fuel cell model to form an objective function satisfying the power demand from the drive cycles. The MPC controller and the rule-based controller are implemented in MATLAB and the powersplit analysis is simulated for all five routes. The results show that the energy management system with the MPC controller optimizes the powertrain configuration efficiently, with preparing for the uphill or downhill, such that the battery SOC stays in its limits and the fuel cell operates in the most efficient range. This ensures operating over different types of drive cycles with the most efficient battery and fuel cell size, hence concluding with the MPC controller outperforming the rule-based one. / Fuel Cell Hybrid Electric Vehicle (FCHEV) är hybridfordon som består av både bränsleceller och batterier som energiomvandlingssystem. Energy ManagementSystem (EMS) spelar en viktig roll i driften av bränslecellshybridsystemet, eftersom det hjälper till att minska systemets vätgasförbrukning. Denna studie undersöker en optimal styralgoritm framtagen i syfte att minska syfte att minska vätgasförbrukningen i bränslecellssystemet. Algoritmen testas på fem olika körcykler, baserade på verkliga Europeiska vägsträckor. Model Predictive Controller (MPC) används för att lösa det optimala styrproblemet, genom att formalisera en framåtblickskontroller med hjälp av dess vikande horisont. Den optimala kontroller jämförs med en konventionell regelbaserad kontroller, genom att analysera hybridsystemet över olika batteri- och bränslecellstorlekar, baserat på den totala väteförbrukningen. Först utvecklas en förenklad systemmodell, som modellerar bränslecellssystemet med avseende på effektivitetskurvan för vätgaskraften och bränslecellseffekten. Batterisystemmodellen med dess State of Charge (SOC) är kopplad till bränslecellsmodellen för att bilda en målfunktion som tillfredsställer kraftbehovet från drivcyklerna. MPC-styrningen och den regelbaserade styrningen är implementerade i matlab och effektdelningsanalysen simuleras för alla fem rutterna. Resultaten visar att energihanteringssystemet medMPC-styrningen optimerar drivlinans konfiguration effektivt, med förberedelser för uppförsbacke eller nedförsbacke, så att batteriets SOC håller sig inom sina gränser och bränslecellen arbetar i mest optimala räckvidden. Detta säkerställer drift över olika typer av körcykler med den mest effektiva batteri- och bränslecellsstorleken, och avslutar därför med att MPC-styrenheten överträffar den regelbaserade.

This thesis addresses the problem of energy management of a hybrid electric power unit for an autonomous vehicle. We introduce, evaluate, and discuss the idea of autonomous-specific energy management strategy. This method is an optimization-based strategy which improves the powertrain fuel economy by exploiting motion planning data. First, to build a firm base for further evaluations, we will develop a high-fidelity system-level model for our case study using MATLAB/Simulink. This model mostly concerns about energy-related aspects of the powertrain and the vehicle. We will derive and implement the equations for each of the model subsystems. We derive model parameters using available data in the literature or online. Evaluation of the developed model shows acceptable conformity with the actual dynamometer data. We will use this model to replace the built-in rule-based logic with the proposed strategy and assess the performance.\par Second, since we are considering an optimization-based approach, we will develop a novel convex representation of the vehicle and powertrain model. This translates to reformulating the model equations using convex functions. Consequently, we will express the fuel-efficient energy management problem as the convex optimization problem. We will solve the optimization problem using dedicated numerical solvers. Extracting the control inputs using this approach and applying them on the high-fidelity model provides similar results to dynamic programming in terms of fuel consumption but in substantially less amount of time. This will act as a pivot for the subsequent real-time analysis.\par Third, we will perform a proof-of-concept for the autonomous-specific energy management strategy. We implement an optimization-based path and trajectory planning for a vehicle in the simplified driving scenario of a racing track. Accordingly, we use motion planning data to obtain the energy management strategy by solving an optimization problem. We will let the vehicle to travel around the circuit with the ability to perceive and plan up to an observable horizon using the receding horizon approach. Developed approach for energy management strategy shows a substantial reduction in the fuel consumption of the high-fidelity model, compared to the rule-based controller. / Thesis / Master of Science in Mechanical Engineering (MSME) / The automotive industry is on the verge of groundbreaking transformations as a result of electrification and autonomous driving. Electrified autonomous car of the future is sustainable, energy-efficient, more convenient, and safer. In addition to the advantages of electrification and autonomous driving individually, the intersection and interaction of these mainstreams provide new opportunities for further improvements on the vehicles. Autonomous cars generate an unprecedented amount of real-time data due to excessive use of perception sensors and processing units. This thesis considers the case of an autonomous hybrid electric vehicle and presents the novel idea of autonomous-specific energy management strategy. Specifically, this thesis is a proof-of-concept, a trial to exploit the motion planning data for a self-driving car to improve the fuel economy of the hybrid electric power unit by adopting a more efficient energy management strategy. With the ever-increasing number of autonomous hybrid electric vehicles, particularly in the self-driving fleets, the presented method shows an extremely promising potential to reduce the fuel consumption of these vehicles.

In this thesis, we explore the potential of financing a fast charging system with energy storage by delivering ancillary services from the energy storage in an optimal way. Specifically, a system delivering frequency regulation services FCR-D Up and FCR-D Down in combination with energy arbitrage trading is considered. An optimization model is developed that could be implemented operationally and then used in a Monte-Carlo simulation to estimate the net present value of the system for four identified cases at three different energy market price scenarios. The main modeling approach is to formulate the system as a state-space model serving as the foundation for model predictive control, with the delay between decision and delivery of the frequency regulation services incorporated as a part of the system state. The optimization of the system is implemented using a dynamic programming approach with a time horizon of 48h, where the choice of admissible controls is optimized for computational efficiency. The result shows that the system could profitable under optimal operation, but it is heavily dependent on the size of the grid connection, future price levels for ancillary services, and the nature of fast-charging demand. As such, the business case and profitability should be evaluated with a specific use case in mind. The developed model showed relatively good computational efficiency for operational implementations with a run time for one iteration of the optimization problem of 15 seconds. The model could therefore be used as the foundation for future research within the specific field and for similar control problems considering delayed controls and stochastic demand. Several proposed improvements and suggested areas of future research are proposed. / I den här uppsatsen utforskar vi huruvida det är finansiellt lönsamt att leverera snabbladdning från ett energilager samtidigt som energilagret används för att leverera systemtjänster på ett optimalt sätt. Mer specifikt undersöks ett potentiellt system som levererar frekvensregleringstjänsterna FCR-D Up och FCR-D Down samt energiarbitragehandel. Vi utvecklar en optimeringsmodell som kan implementeras i ett fysiskt system och använder sedan modellen i en Monte-Carlo-simulering för att estimera nuvärdet av fyra olika systemkonfigurationer för tre olika prisscenarion. Den huvudsakliga modelleringsmetoden är att formulera systemet som en tillstånds-rum modell, som sedan används som grund för modellprediktiv styrning, där fördröjningen mellan beslut och leverans av frekvensregleringstjänster inkluderas som en del av systemets tillstånd. Optimeringen av systemet implementeras med en dynamisk programmeringsmetodik med en tidsram på 48 timmar, där valet av tillåtna kontroller optimeras för beräkningseffektivitet. Resultatet visar att systemet kan vara lönsamt under optimal drift, men det är starkt beroende av storleken på nätanslutningen, framtida prisnivåer för systemtjänster och typen av snabbladdningsbehovet. Därför bör lönsamheten utvärderas för varje specifikt fall. Den utvecklade modellen visade relativt god beräkningseffektivitet för praktiskt implementation med en körtid för en enskilt iteration på 15 sekunder. Modellen kan därför användas som grund för framtida forskning inom området och för liknande problem inom optimal styrteori som involverar fördröjda kontroller och stokastisk efterfrågan. Flera föreslagna förbättringar och områden för framtida forskning föreslås.

Model Predictive Control (MPC) is a control strategy based on optimization that handles system constraints explicitly, making it a popular feedback control method in real industrial processes. However, designing this control policy is an expensive operation since an explicit model of the process is required when re-tuning the controller. Another common practical challenge is that not all states are available, which calls for an observer in order to estimate the states, and imposes additional challenges such as satisfying the constraints and conditions that follow. This thesis attempts to address these challenges by extending the novel Adversarial Adaptive Model Predictive Control (AAMPC) algorithm with output-feedback for linear plants without explicit identification. The AAMPC algorithm is an adaptive MPC framework, where results from an adversarial Multi-Armed Bandit (MAB) are applied to a basic model predictive control formulation. The algorithm of the project, Adversarial Adaptive Output-Feedback Model Predictive Control (AAOFMPC), is derived by extending the standard MPC formulation with output-feedback, i.e, to an Output-Feedback Model Predictive Control (OFMPC) scheme, where a Kalman filter is implemented as the observer. Furthermore, the control performance of the extended algorithm is demonstrated with the problem of driving the state to a given reference, in which the performance is evaluated in terms of regret, state estimation errors, and how well the states track their given reference. Experiments are conducted on two discrete-time Linear Time- Invariant (LTI) systems, a second order system and a third order system, that are perturbed with different noise sequences. It is shown that the AAOFMPC performance satisfies the given theoretical bounds and constraints despite larger perturbations. However, it is also shown that the algorithm is not very robust against noise since offsets from the reference values for the state trajectories are observed. Furthermore, there are several tuning parameters of AAOFMPC that need further investigation for optimal performance. / Modell Prediktiv Reglering (MPC) är en optimeringsbaserad reglertekniksmetod som hanterar processbegränsingar på ett systematiskt sätt, vilket gör den till en populär metod inom återkopplad reglering i processindustrin. Denna metod medför dock höga beräkningskostnader eftersom det krävs en explicit modell varje gång regulatorn justeras online. I praktiken är det också vanligt att alla tillståndsvariabler inte är tillgängliga, vilket kräver en observatör för att rekonstruera alla tillståndsvariabler. Detta leder till fler utmaningar som att uppfylla ytterligare systembegränsingar och villkor som följer. Detta projekt adresserar dessa utmaningar genom att förlänga den nya algoritmen Adversarial Adaptiv Modell Prediktiv Reglering (AAMPC) med output-feedback för linjära system utan explicit modellidentifiering. AAMPC-algoritmen är en adaptiv reglerstrategi där resultat från en adversarial multiarmed bandit (MAB) appliceras i en standard MPC-formulering. Denna MPC-formulering är förlängd med output-feedback dvs. Output-Feedback Modell Predktiv Reglering (OFMPC) där ett Kalman filter är implementerad som en observatör och resulterar i projektets algoritm: Adversarial Adaptiv Output- Feedback Modell Prediktiv Reglering (AAOFMPC). Vidare demonstreras den utökade algoritmens prestanda med problemet att driva tillståndsvariablerna till ett givet referensvärde, där prestandan evalueras i termer av regret, skattningsfel och hur väl tillståndsvariablerna följer de givna referensvärdena. Experiment utförs på två tidsdiskreta tidsinvarianta (LTI) system, ett andraordningssystem och ett tredjeordningssystem, som är perturberade med olika värden av brus. Resultaten visar att AAOFMPC:s prestanda uppfyller de givna teoretiska begränsningarna trots större störningar. Det visar sig dock att algoritmen inte är särskilt robust mot brus eftersom det sker avvikelser från de givna referensvärdena för tillståndsvariablerna. Dessutom finns det flera parametrar i algoritmen som kräver ytterligare utredningar för optimal prestanda.