Search

Sustainable energy systems rely on a wide range of energy sources, where an integral part is to use the available energy as efficiently as possible. District energy systems are considered a key factor as an efficient way of distributing heat and cold within urban areas and facilitating the utilization of renewable energy sources and heat recovery from, e.g., industrial plants and data centers. To achieve high-performing control and an increased understanding of the district energy system, a dynamic model of the process can be used. However, a city-scale, automatically generated, and updated model that can be used for the whole lifecycle of the plant remains a challenging problem. Large-scale physics-based models are sometimes used for planning and validation, but using the models for optimization and control, long-term simulation, or running a high number of simulation scenarios can be computationally prohibitive. In the thesis, the physics of the district energy grid is presented along with modeling, simulation, and control methods with the goal of increasing the computational efficiency and flexibility of the methods. The grid is described using graph theoretical concepts and a linear parameter-varying state-space model representation, along with an introduction to reduced-order models, heat load prediction, Gaussian process models, and feedback control for district energy systems. The main contribution of the thesis is the seven papers. Experiences, challenges, and possible methods to address the presented problems are given in the first paper of the thesis. In the second paper a method for the prediction of heat load for buildings is presented, followed by a machine learning-based method of modeling the thermal dynamics in a district heating pipe. Next, a method for reduced order modeling of district energy grids using graph theoretical methods and spectral clustering is presented in two papers. The sixth paper suggests an integrated approach to spatial and en

The purpose of this licentiate thesis is to research a few key concepts which are important in the development of high efficiency electric drives. The selected research method is quantitative, the raw data have been collected from real world measurements, both from powerlab and from train. The data is analysed and the results are presented in the thesis and associated papers. The paper in this thesis describes and evaluates the impact of SiC MOSFET inverters on the traction drive. Furthermore, the papers go into depth about losses within the traction drive and how the electromagnetic noise depends on both the inverter switching frequency and pulse width modulation method. In the papers, results from low fidelity models of both losses and noise are presented, these models offer greater insight into the mechanism behind both losses and electromagnetic noise. Understanding how the motor harmonic losses depends on the current distribution in conductors, can contribute to improved designs of both motors and inverters. An understanding of how electromagnetic noise increases when voltage harmonics coincide with the mechanical modes of resonance of the electric motor, can further contribute to improved designs of both inverters and motors. The thesis concludes that when SiC based MOSFETs inverters makes its debut into everyday trains, we can expect significant increases in energy efficiency as well as significantly reductions in electromagnetic noise.

The purpose of this licentiate thesis is to research a few key concepts which are important in the development of high efficiency electric drives. The selected research method is quantitative, the raw data have been collected from real world measurements, both from powerlab and from train. The data is analysed and the results are presented in the thesis and associated papers. The paper in this thesis describes and evaluates the impact of SiC MOSFET inverters on the traction drive. Furthermore, the papers go into depth about losses within the traction drive and how the electromagnetic noise depends on both the inverter switching frequency and pulse width modulation method. In the papers, results from low fidelity models of both losses and noise are presented, these models offer greater insight into the mechanism behind both losses and electromagnetic noise. Understanding how the motor harmonic losses depends on the current distribution in conductors, can contribute to improved designs of both motors and inverters. An understanding of how electromagnetic noise increases when voltage harmonics coincide with the mechanical modes of resonance of the electric motor, can further contribute to improved designs of both inverters and motors. The thesis concludes that when SiC based MOSFETs inverters makes its debut into everyday trains, we can expect significant increases in energy efficiency as well as significantly reductions in electromagnetic noise.

This paper proposes a novel placement technique based on the improved golden jackal optimization (IGJO) algorithm for multiple capacitor banks (CBs) and multi-type DGs in a distribution network considering single and multi-objective problems. The proposed algorithm incorporates memory-based equations and random walk strategy to enhance the performance of the recent golden jackal optimization in terms of accuracy and convergence speed. The optimization problem is formulated as a weighted multi-objective that seeks to enhance the voltage profiles, boost stability, and minimize the total active power loss. An index named reactive loss sensitivity (QLSI) is also employed with the developed IGJO to identify the candidate nodes for the DGs and CBs installation to reduce the search space of the optimization algorithm. The robustness of the developed IGJO algorithm is evaluated through the CEC 2020 benchmark functions, and a comparison study is conducted with the original GJO and the other nine fresh competitors using various statistical tests to confirm its dominance and superiority. Then, the proposed IGJO is implemented in single and multi-objectives for the optimal deployment of multiple CBs individually and simultaneously with multiple DGs with different operating modes to enhance the performance of the IEEE 69-bus radial distribution system (RDS). The fetched outcomes are compared with the original GJO, weevil optimizer algorithm (WeevilOA), skill optimization algorithm (SOA), and Tasmanian devil optimization (TDO) to further measure its efficacy using different statistical tests. The IGJO algorithm is also applied to deploy multiple DGs for the IEEE 118-bus RDS with the aim of minimizing active loss. The simulation findings affirmed that the proposed IGJO technique beats the other rivals in all investigated situations, qualifying for the optimal inclusion of DGs in the presence of generation and demand uncertainties. Specifically, the integration of three units of CB

At the beginning of this century, Hegselmann and Krause proposed a dynamical model for opinion formation that is referred to as the Bounded Confidence Opinion Dynamics (BCOD) model and that has since attracted a wide interest from different research communities. The model can be viewed as a dynamic network, in which each agent is endowed with a state variable representing an opinion and two agents interact if the distance between their opinions does not exceed a constant confidence bound. This relation of instantaneous proximity between the opinions naturally induces a dynamic interaction graph. At each stage of the opinion iteration, all agents synchronously update their opinion to the average of all opinions that belong to the neighbors in the interaction graph.BCOD models exhibit a broad variety of phenomena that cannot be studied by traditional methods, and their analysis has enriched the systems and control field with a number of novel mathematical tools. This fact, together with the existence of an extensive literature on the topic scattered across different fields, calls for a systematic presentation of the existing results on this class of dynamic networks. The aim of this survey is to provide an overview of BCOD models with time-synchronous interactions, with possibly asymmetric and heterogeneous confidence bounds. Conditions on the different classes of BCOD which ensure the convergence (in finite time or asymptotically) of the opinions are discussed, and the possible structures of the terminal opinions are described. The numerous phenomena highlighted in the literature from numerical studies, e.g., the characterization of steady state behaviors and the sensitivity to confidence thresholds, are also reviewed. Finally, some recent modifications and applications of BCOD models are discussed, and suggestions of directions for future research are provided., Funding Agencies|Ministero dellUniversit e Ricerca [2022K8EZBW]

The purpose of this licentiate thesis is to research a few key concepts which are important in the development of high efficiency electric drives. The selected research method is quantitative, the raw data have been collected from real world measurements, both from powerlab and from train. The data is analysed and the results are presented in the thesis and associated papers. The paper in this thesis describes and evaluates the impact of SiC MOSFET inverters on the traction drive. Furthermore, the papers go into depth about losses within the traction drive and how the electromagnetic noise depends on both the inverter switching frequency and pulse width modulation method. In the papers, results from low fidelity models of both losses and noise are presented, these models offer greater insight into the mechanism behind both losses and electromagnetic noise. Understanding how the motor harmonic losses depends on the current distribution in conductors, can contribute to improved designs of both motors and inverters. An understanding of how electromagnetic noise increases when voltage harmonics coincide with the mechanical modes of resonance of the electric motor, can further contribute to improved designs of both inverters and motors. The thesis concludes that when SiC based MOSFETs inverters makes its debut into everyday trains, we can expect significant increases in energy efficiency as well as significantly reductions in electromagnetic noise.

The purpose of this licentiate thesis is to research a few key concepts which are important in the development of high efficiency electric drives. The selected research method is quantitative, the raw data have been collected from real world measurements, both from powerlab and from train. The data is analysed and the results are presented in the thesis and associated papers. The paper in this thesis describes and evaluates the impact of SiC MOSFET inverters on the traction drive. Furthermore, the papers go into depth about losses within the traction drive and how the electromagnetic noise depends on both the inverter switching frequency and pulse width modulation method. In the papers, results from low fidelity models of both losses and noise are presented, these models offer greater insight into the mechanism behind both losses and electromagnetic noise. Understanding how the motor harmonic losses depends on the current distribution in conductors, can contribute to improved designs of both motors and inverters. An understanding of how electromagnetic noise increases when voltage harmonics coincide with the mechanical modes of resonance of the electric motor, can further contribute to improved designs of both inverters and motors. The thesis concludes that when SiC based MOSFETs inverters makes its debut into everyday trains, we can expect significant increases in energy efficiency as well as significantly reductions in electromagnetic noise.

The pure time delay operator is considered in continuous and discrete time under the assumption of the input signal being integrable (summable) with square. Then the input and the output signals are uniquely given by their Laguerre spectra. It is shown that a discrete convolution operator with polynomial Markov parameters constitutes a common description of the delay operator in the continuous and discrete case. Closed-form expressions for the delay value in terms of the output and input Laguerre spectra are derived. The expressions hold for any feasible value of the Laguerre parameter and can be utilized for e.g. building time-delay estimators that allow for non-persistent input. A simulation example is provided to illustrate the principle of Laguerre-domain time-delay modeling and analysis with perfect disturbance rejection.

The pure time delay operator is considered in continuous and discrete time under the assumption of the input signal being integrable (summable) with square. Then the input and the output signals are uniquely given by their Laguerre spectra. It is shown that a discrete convolution operator with polynomial Markov parameters constitutes a common description of the delay operator in the continuous and discrete case. Closed-form expressions for the delay value in terms of the output and input Laguerre spectra are derived. The expressions hold for any feasible value of the Laguerre parameter and can be utilized for e.g. building time-delay estimators that allow for non-persistent input. A simulation example is provided to illustrate the principle of Laguerre-domain time-delay modeling and analysis with perfect disturbance rejection.

Neural networks are ubiquitous components of Machine Learning (ML) algorithms. However, training them is challenging due to problems associated with exploding and vanishing loss-gradients. Gradient clipping is shown to effectively combat both the vanishing gradients and the exploding gradients problems. As the name suggests, gradients are clipped in order to prevent large updates. At the same time, very small neural network weights are updated using larger step-sizes. Although widely used in practice, there is very little theory surrounding clipping. In this paper, we analyze two popular gradient clipping techniques-the classic norm-based gradient clipping method and the adaptive gradient clipping technique. We prove that gradient clipping ensures numerical stability with very high probability. Further, clipping based stochastic gradient descent converges to a set of neural network weights that minimizes the average scaled training loss in a local sense. The averaging is with respect to the distribution that generated the training data. The scaling is a consequence of gradient clipping. We use tools from the theory of dynamical systems for the presented analysis.

Energy management within microgrids under the presence of large number of renewables such as photovoltaics is complicated due to uncertainties involved. Randomness in energy production and consumption make both the prediction and optimality of exchanges challenging. In this paper, we evaluate the impact of uncertainties on optimality of different robust energy exchange strategies. To address the problem, we present AIROBE, a data-driven system that uses machine-learning-based predictions of energy supply and demand as input to calculate robust energy exchange schedules using a multiband robust optimization approach to protect from deviations. AIROBE allows the decision maker to tradeoff robustness with stability of the system and energy costs. Our evaluation shows, how AIROBE can deal effectively with asymmetric deviations and how better prediction methods can reduce both the operational cost while at the same time may lead to increased operational stability of the system.

The uncertainty in the prediction calculated using the delta method for an over-parameterized (parametric) black-box model is shown to be larger or equal to the uncertainty in the prediction of a canonical (minimal) model. Equality holds if the additional parameters of the overparameterized model do not add flexibility to the model. As a conclusion, for an overparameterized black-box model, the calculated uncertainty in the prediction by the delta method is not underestimated. The results are shown analytically and are validated in a simulation experiment where the relationship between the normalized traction force and the wheel slip of a car is modelled using e.g., a neural network.

Suppose data-driven black-box models, e.g., neural networks, should be used as components in safety-critical systems such as autonomous vehicles. In that case, knowing how uncertain they are in their predictions is crucial. However, this needs to be provided for standard formulations of neural networks. Hence, this thesis aims to develop a method that can, out-of-the-box, extend the standard formulations to include uncertainty in the prediction. The proposed method in the thesis is based on a local linear approximation, using a two-step linearization to quantify the uncertainty in the prediction from the neural network. First, the posterior distribution of the neural network parameters is approximated using a Gaussian distribution. The mean of the distribution is at the maximum a posteriori estimate of the parameters, and the covariance is estimated using the shape of the likelihood function in the vicinity of the estimated parameters. The second linearization is used to propagate the uncertainty in the parameters to uncertainty in the model’s output. Hence, to create a linear approximation of the nonlinear model that a neural network is. The first part of the thesis considers regression problems with examples of road-friction experiments using simulated and experimentally collected data. For the model-order selection problem, it is shown that the method does not under-estimate the uncertainty in the prediction of overparametrized models. The second part of the thesis considers classification problems. The concept of calibration of the uncertainty, i.e., how reliable the uncertainty is and how close it resembles the true uncertainty, is considered. The proposed method is shown to create calibrated estimates of the uncertainty, evaluated on classical image data sets. From a computational perspective, the thesis proposes a recursive update of the parameter covariance, enhancing the method’s viability. Furthermore, it shows how quantified uncertainty can improve the, Funding: The research work in this thesis has been supported by the Sweden's Innovation Agency, Vinnova, through project iQDeep (project number 2018-02700).

Using sensors to observe real-world systems is important in many applications. A typical use case is target tracking, where sensor measurements are used to compute estimates of targets. Two of the main purposes of the estimates are to enhance situational awareness and facilitate decision-making. Hence, the estimation quality is crucial. By utilizing multiple sensors, the estimation quality can be further improved. Here, the focus is on target tracking in decentralized sensor networks, where multiple agents estimate a common set of targets. In a decentralized context, measurements undergo local preprocessing at the agent level, resulting in local estimates. These estimates are subsequently shared among the agents for estimate fusion. Sharing information leads to correlations between estimates, which in decentralized sensor networks are often unknown. In addition, there are situations where the communication capacity is constrained, such that the shared information needs to be reduced. This thesis addresses two aspects of decentralized target tracking: (i) fusion of estimates with unknown correlations; and (ii) handling of constrained communication resources. Decentralized sensor networks have unknown correlations because it is typically impossible to keep track of dependencies between estimates. A common approach in this case is to use conservative estimators, which can ensure that the true uncertainty of an estimate is not underestimated. This class of estimators is pursued here. A significant part of the thesis is dedicated to the widely-used conservative method known as covariance intersection (CI), while also describing and deriving alternative methods for CI. One major result related to aspect (i) is the conservative linear unbiased estimator (CLUE), which is proposed as a general framework for optimal conservative estimation. It is shown that several existing methods, including CI, are optimal CLUEs under different conditions. A decentralized sensor network a, Funding: VINNOVA and Saab AB through the LINK-SIC Competence Center.

This paper introduces a disturbance-parametrized (DP) robust lattice-based motion-planning framework for nonlinear systems affected by bounded disturbances. A key idea in this work is to rigorously exploit the available knowledge about the disturbance, starting already offline at the time when a library of DP motion primitives is computed and ending not before the motion has been executed online. Given an up-to-date-estimate of the disturbance, the lattice-based motion planner performs a graph search online, to non-conservatively compute a disturbance aware optimal motion plan with formally motivated margins to obstacles. This is done utilizing the DP motion primitives, around which tubes are generated utilizing a suitably designed robust controller. The sizes of the tubes are dependent on the upper bounds of the disturbance appearing in the error between the actual system trajectory and the DP nominal trajectory, which in turn along with the overall optimality of the plan is dependant on the user-selected resolution of the available disturbance estimates. Increasing the resolution of the disturbance parameter results in smaller sizes of tubes around the motion primitives and can significantly reduce the conservativeness compared to traditional approaches, thus increasing the performance of the computed motion plans. The proposed strategy is implemented on an Euler-Lagrange-based ship model which is affected by a significant wind disturbance and the efficiency of the strategy is validated through a suitable simulation example., Funding: ELLIIT

A new class of iterated linearization-based nonlinear filters, dubbed dynamically iterated filters, is presented. Contrary to regular iterated filters such as the iterated extended Kalman filter (IEKF), iterated unscented Kalman filter (IUKF) and iterated posterior linearization filter (IPLF), dynamically iterated filters also take nonlinearities in the transition model into account. The general filtering algorithm is shown to essentially be a (locally over one time step) iterated Rauch-Tung-Striebel smoother. Three distinct versions of the dynamically iterated filters are especially investigated: analogues to the IEKF, IUKF and IPLF. The developed algorithms are evaluated on 25 different noise configurations of a tracking problem with a nonlinear transition model and linear measurement model, a scenario where conventional iterated filters are not useful. Even in this “simple” scenario, the dynamically iterated filters are shown to have superior root mean-squared error performance as compared with their respective baselines, the EKF and UKF. Particularly, even though the EKF diverges in 22 out of 25 configurations, the dynamically iterated EKF remains stable in 20 out of 25 scenarios, only diverging under high noise.

Energy management within microgrids under the presence of large number of renewables such as photovoltaics is complicated due to uncertainties involved. Randomness in energy production and consumption make both the prediction and optimality of exchanges challenging. In this paper, we evaluate the impact of uncertainties on optimality of different robust energy exchange strategies. To address the problem, we present AIROBE, a data-driven system that uses machine-learning-based predictions of energy supply and demand as input to calculate robust energy exchange schedules using a multiband robust optimization approach to protect from deviations. AIROBE allows the decision maker to tradeoff robustness with stability of the system and energy costs. Our evaluation shows, how AIROBE can deal effectively with asymmetric deviations and how better prediction methods can reduce both the operational cost while at the same time may lead to increased operational stability of the system.

Neural networks are ubiquitous components of Machine Learning (ML) algorithms. However, training them is challenging due to problems associated with exploding and vanishing loss-gradients. Gradient clipping is shown to effectively combat both the vanishing gradients and the exploding gradients problems. As the name suggests, gradients are clipped in order to prevent large updates. At the same time, very small neural network weights are updated using larger step-sizes. Although widely used in practice, there is very little theory surrounding clipping. In this paper, we analyze two popular gradient clipping techniques-the classic norm-based gradient clipping method and the adaptive gradient clipping technique. We prove that gradient clipping ensures numerical stability with very high probability. Further, clipping based stochastic gradient descent converges to a set of neural network weights that minimizes the average scaled training loss in a local sense. The averaging is with respect to the distribution that generated the training data. The scaling is a consequence of gradient clipping. We use tools from the theory of dynamical systems for the presented analysis.

Decomposition of organic matter controls the flow of carbon and nutrients in terrestrial and aquatic ecosystems. Several kinetic laws have been proposed to describe decomposition rates, but they neglect adaptation of the microbial decomposer to environmental conditions. Here we formalise decomposition as an optimal control problem by assuming that microorganisms regulate the uptake rate of a substrate to maximise their growth over the period of decomposition. The result is an optimal control problem consisting of two differential equations and auxiliary conditions that determine the optimal value of the control variable (the uptake rate), the remaining substrate at any given time, and the optimal completion time. This problem serves as a case study to illustrate the solution of differential equations and optimal control problems for students in undergraduate courses. The mathematical analysis of the problem requires rewriting the differential equations in reverse time along with the solution of a nonhomogeneous linear first order differential equation. We then return to modelling with some biologically motivated questions about how the parameters of the model representing environmental conditions and microbial functional traits affect the outcome. Finally, we discuss alternative ways to use the material with students.

Suppose data-driven black-box models, e.g., neural networks, should be used as components in safety-critical systems such as autonomous vehicles. In that case, knowing how uncertain they are in their predictions is crucial. However, this needs to be provided for standard formulations of neural networks. Hence, this thesis aims to develop a method that can, out-of-the-box, extend the standard formulations to include uncertainty in the prediction. The proposed method in the thesis is based on a local linear approximation, using a two-step linearization to quantify the uncertainty in the prediction from the neural network. First, the posterior distribution of the neural network parameters is approximated using a Gaussian distribution. The mean of the distribution is at the maximum a posteriori estimate of the parameters, and the covariance is estimated using the shape of the likelihood function in the vicinity of the estimated parameters. The second linearization is used to propagate the uncertainty in the parameters to uncertainty in the model’s output. Hence, to create a linear approximation of the nonlinear model that a neural network is. The first part of the thesis considers regression problems with examples of road-friction experiments using simulated and experimentally collected data. For the model-order selection problem, it is shown that the method does not under-estimate the uncertainty in the prediction of overparametrized models. The second part of the thesis considers classification problems. The concept of calibration of the uncertainty, i.e., how reliable the uncertainty is and how close it resembles the true uncertainty, is considered. The proposed method is shown to create calibrated estimates of the uncertainty, evaluated on classical image data sets. From a computational perspective, the thesis proposes a recursive update of the parameter covariance, enhancing the method’s viability. Furthermore, it shows how quantified uncertainty can improve the, Funding: The research work in this thesis has been supported by the Sweden's Innovation Agency, Vinnova, through project iQDeep (project number 2018-02700).

In this thesis, we study the related problems of reinforcement learning and optimal adaptive control, specialized to specific classes of stochastic and structured dynamical systems. By stochastic, we mean systems that are unknown to the decision maker and evolve according to some probabilistic law. By structured, we mean that they are restricted in some known way, e.g., they belong to a specific model class or must obey a set of known constraints. The objective in both problems is the design of an optimal algorithm, i.e., one that maximizes a certain performance metric. Because of the stochasticity, the algorithm faces an exploration-exploitation dilemma, where it must balance collecting information from the system and leveraging existing information to choose the best action or input. This trade-off is best captured by the notion of regret, defined as the difference between the performance of the algorithm and an oracle which has full knowledge of the system. In the first part of the thesis, we investigate systems that can be modeled as Markov Decision Processes (MDPs) and derive general asymptotic and problem-specific regret lower bounds for ergodic and deterministic MDPs. We make these bounds explicit for MDPs that: i) are ergodic and unstructured, ii) have Lipschitz transitions and rewards, and iii) are deterministic and satisfy a decoupling property. Furthermore, we propose Directed Exploration Leaning (DEL), an algorithm that is valid for any ergodic MDP with any structure and whose regret upper bound matches the associated regret lower bounds, thus being truly optimal. For this algorithm, we present theoretical regret guarantees as well as a numerical demonstration that verifies its ability to exploit the underlying structure. In the second part, we study systems with uncertain linear dynamics and which are subject to additive disturbances as well as state and input constraints. We develop Self-Tuning Tube-based Model Predictive Control (STTMPC), an adaptive, QC 20231003

Autonomous Vehicle (AV) technology promises safer, greener, and more efficient means of transportation for everyone. AVs are expected to have their first big impact in closed environments, such as mining areas, ports, and construction sites, where Heavy-Duty Vehicles (HDVs) operate. This thesis addresses lateral motion control for autonomous HDVs using Model Predictive Control (MPC). Lateral control for HDVs still has many open questions to be addressed, in particular, precise path tracking while ensuring a smooth, comfortable, and stable ride, coping with both external and internal disturbances, and adapting to different vehicles and conditions. To address these challenges, a comprehensive control module architecture is designed to adapt seamlessly to different vehicle types and interface with various planning and localization modules. Furthermore, it is designed to address system delays, maintain certain error bounds, and respect actuation constraints. This thesis presents the Reference Aware MPC (RA-MPC) for autonomous vehicles. This controller is iteratively improved throughout the thesis. The RA-MPC introduces a method to systematically handle references generated by motion planners which can consider different algorithms and vehicle models from the controller. The controller uses the linear time-varying MPC framework and considers control input rate and acceleration constraints to account for steering limitations. Furthermore, multiple models and control inputs are considered throughout the thesis. Ultimately, curvature acceleration is used as the control input, which together with stability ingredients, allows for stability guarantees under certain conditions via Lyapunov techniques. MPC is highly dependent on the prediction model used. This thesis proposes and compares different models. First, an offline-fitted, vehicle-specific nonlinear curvature response function is proposed and integrated into the kinematic bicycle model. The curvature response function, Teknik för autonoma fordon lovar säkrare, grönare och effektivare transportmedel för alla. Autonoma fordon förväntas få sin första stora inverkan i inhägnade områden, såsom gruvområden, hamnar och byggarbetsplatser, där tunga fordon är i drift. Denna avhandling behandlar lateral rörelsereglering för autonoma tunga fordon med hjälp av modellprediktiv reglering (Model Predictive Control, MPC). Lateral rörelsereglering för tunga fordon har fortfarande flera öppna utmaningar, i synnerhet gällande precis banföljning som garanterar en mjuk, komfortabel och stabil resa, hanterar både externa och interna störningar och anpassar sig till olika fordon och förhållanden. För att möta dessa utmaningar är en omfattande regleringsarkitektur utformad för att sömlöst anpassa sig till olika fordonstyper och gränssnitt mot olika planering- och lokaliseringsmoduler. Därutöver är arkitekturen utformad för att hantera systemfördröjningar, bibehålla särskilda felmarginaler och respektera styrdons begränsningar. Denna avhandling presenterar den Referensmedvetna modellprediktiva regulatorn (Reference Aware MPC, RA-MPC) för autonoma fordon. Denna regulator är iterativt för-bättrad genom hela avhandlingen. Den referensmedvetna modellprediktiva regulatorn introducerar en metod för att systematiskt hantera referenssignaler genererade av rörelseplanerare som beaktar andra typer av algoritmer och fordonsmodeller än regulatorn. Regulatorn använder det linjärt tidsvarierande MPC-ramverket och beaktar begränsningar på styrsignalsförändringar och accelerationer för att ta hänsyn till begränsningar på styrningen. Vidare beaktas flertalet modeller och styrsignaler genom hela avhandlingen. I slutändan används kurvaturacceleration som styrsignal, vilket tillsammans med stabilitetsingredienser möjliggör stabilitetsgarantier under särskilda förhållanden via Lyapunov-tekniker. MPC är starkt beroende av den prediktionsmodell som används. Denna avhandling föreslår och jämför ett flertal olika modeller. Först, QC 20231003

Using sensors to observe real-world systems is important in many applications. A typical use case is target tracking, where sensor measurements are used to compute estimates of targets. Two of the main purposes of the estimates are to enhance situational awareness and facilitate decision-making. Hence, the estimation quality is crucial. By utilizing multiple sensors, the estimation quality can be further improved. Here, the focus is on target tracking in decentralized sensor networks, where multiple agents estimate a common set of targets. In a decentralized context, measurements undergo local preprocessing at the agent level, resulting in local estimates. These estimates are subsequently shared among the agents for estimate fusion. Sharing information leads to correlations between estimates, which in decentralized sensor networks are often unknown. In addition, there are situations where the communication capacity is constrained, such that the shared information needs to be reduced. This thesis addresses two aspects of decentralized target tracking: (i) fusion of estimates with unknown correlations; and (ii) handling of constrained communication resources. Decentralized sensor networks have unknown correlations because it is typically impossible to keep track of dependencies between estimates. A common approach in this case is to use conservative estimators, which can ensure that the true uncertainty of an estimate is not underestimated. This class of estimators is pursued here. A significant part of the thesis is dedicated to the widely-used conservative method known as covariance intersection (CI), while also describing and deriving alternative methods for CI. One major result related to aspect (i) is the conservative linear unbiased estimator (CLUE), which is proposed as a general framework for optimal conservative estimation. It is shown that several existing methods, including CI, are optimal CLUEs under different conditions. A decentralized sensor network a, Funding: VINNOVA and Saab AB through the LINK-SIC Competence Center.

Cyberattacks against critical infrastructures pose a serious threat to society, as they can have devastating consequences on the economy, security, or public health. These infrastructures rely on a large network of cyber components, such as sensors, controllers, computers, and communication devices, to monitor and control their physical processes. An adversary can exploit the vulnerabilities in these cyber components to gain access to the system and manipulate its behavior or functionality. This thesis proposes methods that can be employed as a first line of defense against such attacks for Cyber-Physical Systems. In the first part of the thesis, we consider how uninformed attackers can learn to attack a Cyber-Physical System by eavesdropping through the cyber component. By learning to manipulate the plant, the attacker could figure out how to destroy the physical system before it is too late or completely take it over without raising any alarms. Stopping the attacker at the learning stage would force the attacker to act obliviously, increasing the chances of detecting them. We analyze how homomorphic encryption, a technique that allows computation on encrypted data, hinders an attacker's learning process and reduces its capabilities to attack the system. Specifically, we show that an attacker must solve challenging lattice problems to find attacks that are difficult to detect. Additionally, we show how the detection probability is affected by the attacker's solution to the problems and what parameters of the encryption scheme can be tweaked to increase the detection probability. We also develop a novel method that enables anomaly detection over homomorphically encrypted data without revealing the actual signals to the detector, thereby discouraging attackers from launching attacks on the detector. The detection can be performed using a hypothesis test. However, special care must be taken to ensure that fresh samples are used to detect changes from nominal behavior., Cyberattacker mot kritisk infrastruktur utgör ett allvarligt hot mot samhället, eftersom de kan få förödande konsekvenser för ekonomin, säkerheten eller folkhälsan. Dessa infrastrukturer utgörs ofta av ett stort nätverk av cyberkomponenter, såsom sensorer, styrenheter, datorer och kommunikationsenheter, för att övervaka och styra sina fysiska processer. En angripare kan utnyttja sårbarheterna i dessa cyberkomponenter för att få tillgång till systemet och därefter manipulera dess beteende eller funktionalitet. Denna avhandling behandlar och föreslår metoder som kan användas som en första försvarslinje mot sådana attacker för cyberfysiska system. I den första delen av avhandlingen undersöker vi hur oinformerade angripare kan lära sig att attackera ett cyberfysiskt system genom att avlyssna dem via cyberkomponenten. Genom att lära sig att manipulera det fysiska systemet kan angriparen ta reda på hur man kan förstöra, eller helt ta över det cyberfysiska systemet utan att något alarm går av förrän det är försent. Genom att stoppa angriparen i inlärningsfasen tvingas angriparen att agera mer omedvetet, vilket ökar chanserna att upptäcka dem. Vi analyserar hur homomorf kryptering, vilket är en krypteringsmetod som möjliggör beräkning med krypterad data, hindrar angriparens inlärnings-process och minskar dess förmåga att attackera systemet. Specifikt visar vi att en angripare måste lösa svåra gitterproblem för att hitta svårdetekterade cyberattacker. Dessutom visar vi hur detektionssannolikheten påverkas av hur bra angriparens lösning är och vilka parametrar i krypteringsschemat som kan justeras för att öka sannolikheten att upptäcka anfallet. Vi utvecklar också en ny metod som möjliggör anomalidetektering över homomorft krypterade data, utan att avslöja de faktiska signalerna för detektorn och därmed avskräcka angripare från att attackera detektorn. Detektionen kan utföras med hjälp av ett hypotestest. Dock måste man se till att färska prover används för att upptäcka förän, QC 20231020

Inom reglerteknik har integrationen av maskininlärningsmetoder framträtt som en central strategi för att förbättra prestanda och adaptivitet hos styrsystem. Betydande framsteg har gjorts inom flera viktiga aspekter av reglerkretsen, såsom inlärningsbaserade metoder för systemidentifiering och parameterskattning, filtrering och brusreducering samt reglersyntes. Denna avhandling fördjupar sig i området inlärning för reglerteknik med särskild betoning på inlärningsbaserade regulatorer och identifieringsmetoder. Avhandlingens första del behandlar undersökningen av neuronnätsbaserad Modellprediktiv Reglering (MPC). Olika nätstrukturer studeras, både generella black box-nät och nät som väver in MPC-specifik information i sin struktur. Dessa nät jämförs och utvärderas med avseende på två prestandamått genom experiment på realistiska två- och fyrdimensionella system. Den huvudsakliga nyskapande aspekten är inkluderingen av gradientdata i träningsprocessen, vilket visar sig förbättra noggrannheten av de genererade styrsignalerna. Vidare påvisar de experimentella resultaten att en MPC-informerad nätstruktur leder till förbättrad prestanda när mängden träningsdata är begränsad. Med insikt om vikten av noggranna matematiska modeller av styrsystemet, riktar den andra delen av avhandlingen sitt fokus mot inlärningsbaserade identifieringsmetoder. Denna forskningsgren behandlar karakterisering och modellering av dynamiska system med hjälp av maskininlärning. Avhandlingen bidrar till området genom att introducera kooperativa systemidentifieringsmetoder för att förbättra parameterskattningen. Specifikt utnyttjas verktyg från Optimal Transport för att introducera en ny och mer generell formulering av ramverket Correctional Learning. Detta ramverk är baserat på en mästare-lärlingsmodell, där en expertagent (mästare) observerar och modifierar den insamlade data som används av en lärande agent (lärling), med syftet att förbättra lärlingens skattningsprocess. Genom att formulera correct, In the context of control systems, the integration of machine learning mechanisms has emerged as a key approach for improving performance and adaptability. Notable progress has been made across several aspects of the control loop, including learning-based techniques for system identification and estimation, filtering and denoising, and controller design. This thesis delves into the rapidly expanding domain of learning in control, with a particular focus placed on learning-based controllers and learning-based identification methods. The first part of this thesis is devoted to the investigation of Neural Network approximations of Model Predictive Control (MPC). Model-agnostic neural network structures are compared to networks employing MPC-specific information, and evaluated in terms of two performance metrics. The main novel aspect lies in the incorporation of gradient data in the training process, which is shown to enhance the accuracy of the network generated control inputs. Furthermore, experimental results reveal that MPC-informed networks outperform the agnostic counterparts in scenarios when training data is limited. In acknowledgement of the crucial role accurate system models play in in the control loop, the second part of this thesis lends its focus to learning-based identification methods. This line of work addresses the important task of characterizing and modeling dynamical systems, by introducing cooperative system identification techniques to enhance estimation performance. Specifically, it presents a novel and generalized formulation of the Correctional Learning framework, leveraging tools from Optimal Transport. The correctional learning framework centers around a teacher-student model, where an expert agent (teacher) modifies the sampled data used by the learner agent (student), to improve the student's estimation process. By formulating correctional learning as an optimal transport problem, a more adaptable framework is achieved, better suited for e, VR 2018-03438 projekt 3224

Industrial automation and control systems (IACS) are taking care of our most important infrastructures, providing electricity and clean water, producing medicine and food, along with many other services and products we take for granted. The continuous, safe, and secure operation of such systems are obviously of great importance. Future iterations of IACS will look quite different from the ones we use today. Modular and flexible systems are emerging, powered by technical advances in areas such as artificial intelligence, cloud computing, and motivated by fluctuating market demands and faster innovation cycles. Design strategies for dynamic manufacturing are increasingly being adopted. These advances have a fundamental impact on industrial systems at component as well as architectural level. As a consequence of the changing operational requirements, the methods used for protection of industrial systems must be revisited and strengthened. This for example includes access control, which is one of the fundamental cyber­security mechanisms that is hugely affected by current developments within IACS. The methods currently used are static and coarse-grained and therefore not well suited for dynamic and flexible industrial systems. A transition in security model is required, from implicit trust towards zero-trust, supporting dynamic and fine-grained access control. This PhD thesis discusses access control for IACS in the age of Industry 4.0, focusing on dynamic and flexible manufacturing systems. The solutions pre­sented are applicable at machine-to-machine as well as human-to-machine in­teractions, using a zero-trust strategy. An investigation of the current state of practice for industrial access control is provided as a starting point for the work. Dynamic systems require equally dynamic access control policies, why several approaches on how dynamic access control can be achieved in indus­trial systems are developed and evaluated, covering strategies for policy for­mula, Vi tar för givet att det alltid ska finnas el, rent dricksvatten, mat och läkemedel. Många av våra grundläggande behov tillgodoses tack vare produkter som är beroende av industriella styrsystem. Att skyddas dessa system ifrån störningar är följaktligen ytterst viktigt. Vi är mitt i ett teknikskifte som brukar kallas "Industri 4.0" och som innebär att framtidens industriella system kommer skilja sig avsevärt ifrån dagens. Förän­dringen drivs bland annat av nya krav och förväntningar, exempelvis på ko­rtare tid mellan ide och produktion, möjlighet att anpassa produktionen till snabba marknadsförändringar och tillverkning av individuellt anpassade pro­dukter. Flexibla och skalbara lösningar krävs för att kunna uppfylla dessa krav, till skillnad från dagens system som i allmänhet är utvecklade för massproduk­tion av en specifik produkt. Detta påverkar såväl hur produktionssystemen konstrueras som designen av varje ingående komponent. En konsekvens är att metoderna som används för att skydda dagens system måste anpassas och stärkas för att möta framtidens utmaningar. En grundläggande sådan säkerhetsfunktion är behörighetshanter­ing. Nuvarande behörighetshantering är inte tillräckligt flexibel och därmed dåligt anpassad till morgondagens dynamiska system. I denna doktorsavhandling undersöks behörighetshantering för framtidens industriella system, med fokus på de dynamiska produktionssystem som behövs för att uppfylla kraven kopplade till Industri 4.0. Med utgångspunkt från en enkätundersökning analyseras dagsläget. Förslag på flera olika tillvägagångssätt för dynamisk behörighetshantering presenteras och utvärderas, såväl med avseende på hur sådana regler kan formuleras som på hur de ska kunna upprätthållas., ARRAY, InSecTT

A multiphase machine control implementing vector space decomposition or harmonic plane decomposition for modeling purposes can reconfigure the stator's current distribution to generate magnetomotive force waves with different pole pair numbers simultaneously. The pole pair numbers are an integer multiple of the pole pairs generated by the physical winding configuration. This property can be used for reconfiguring the pole-pair number in real time or increasing the system's torque density. This paper studies the modeling and the impact of symmetrical and asymmetrical winding configurations on the pole-pair number reconfiguration. Furthermore, pole number transition is experimentally verified on a proposed winding configuration., Part of proceedings ISBN 979-8-3503-9996-7QC 20231114

Inom reglerteknik har integrationen av maskininlärningsmetoder framträtt som en central strategi för att förbättra prestanda och adaptivitet hos styrsystem. Betydande framsteg har gjorts inom flera viktiga aspekter av reglerkretsen, såsom inlärningsbaserade metoder för systemidentifiering och parameterskattning, filtrering och brusreducering samt reglersyntes. Denna avhandling fördjupar sig i området inlärning för reglerteknik med särskild betoning på inlärningsbaserade regulatorer och identifieringsmetoder. Avhandlingens första del behandlar undersökningen av neuronnätsbaserad Modellprediktiv Reglering (MPC). Olika nätstrukturer studeras, både generella black box-nät och nät som väver in MPC-specifik information i sin struktur. Dessa nät jämförs och utvärderas med avseende på två prestandamått genom experiment på realistiska två- och fyrdimensionella system. Den huvudsakliga nyskapande aspekten är inkluderingen av gradientdata i träningsprocessen, vilket visar sig förbättra noggrannheten av de genererade styrsignalerna. Vidare påvisar de experimentella resultaten att en MPC-informerad nätstruktur leder till förbättrad prestanda när mängden träningsdata är begränsad. Med insikt om vikten av noggranna matematiska modeller av styrsystemet, riktar den andra delen av avhandlingen sitt fokus mot inlärningsbaserade identifieringsmetoder. Denna forskningsgren behandlar karakterisering och modellering av dynamiska system med hjälp av maskininlärning. Avhandlingen bidrar till området genom att introducera kooperativa systemidentifieringsmetoder för att förbättra parameterskattningen. Specifikt utnyttjas verktyg från Optimal Transport för att introducera en ny och mer generell formulering av ramverket Correctional Learning. Detta ramverk är baserat på en mästare-lärlingsmodell, där en expertagent (mästare) observerar och modifierar den insamlade data som används av en lärande agent (lärling), med syftet att förbättra lärlingens skattningsprocess. Genom att formulera correct, In the context of control systems, the integration of machine learning mechanisms has emerged as a key approach for improving performance and adaptability. Notable progress has been made across several aspects of the control loop, including learning-based techniques for system identification and estimation, filtering and denoising, and controller design. This thesis delves into the rapidly expanding domain of learning in control, with a particular focus placed on learning-based controllers and learning-based identification methods. The first part of this thesis is devoted to the investigation of Neural Network approximations of Model Predictive Control (MPC). Model-agnostic neural network structures are compared to networks employing MPC-specific information, and evaluated in terms of two performance metrics. The main novel aspect lies in the incorporation of gradient data in the training process, which is shown to enhance the accuracy of the network generated control inputs. Furthermore, experimental results reveal that MPC-informed networks outperform the agnostic counterparts in scenarios when training data is limited. In acknowledgement of the crucial role accurate system models play in in the control loop, the second part of this thesis lends its focus to learning-based identification methods. This line of work addresses the important task of characterizing and modeling dynamical systems, by introducing cooperative system identification techniques to enhance estimation performance. Specifically, it presents a novel and generalized formulation of the Correctional Learning framework, leveraging tools from Optimal Transport. The correctional learning framework centers around a teacher-student model, where an expert agent (teacher) modifies the sampled data used by the learner agent (student), to improve the student's estimation process. By formulating correctional learning as an optimal transport problem, a more adaptable framework is achieved, better suited for e, VR 2018-03438 projekt 3224

Cyberattacks against critical infrastructures pose a serious threat to society, as they can have devastating consequences on the economy, security, or public health. These infrastructures rely on a large network of cyber components, such as sensors, controllers, computers, and communication devices, to monitor and control their physical processes. An adversary can exploit the vulnerabilities in these cyber components to gain access to the system and manipulate its behavior or functionality. This thesis proposes methods that can be employed as a first line of defense against such attacks for Cyber-Physical Systems. In the first part of the thesis, we consider how uninformed attackers can learn to attack a Cyber-Physical System by eavesdropping through the cyber component. By learning to manipulate the plant, the attacker could figure out how to destroy the physical system before it is too late or completely take it over without raising any alarms. Stopping the attacker at the learning stage would force the attacker to act obliviously, increasing the chances of detecting them. We analyze how homomorphic encryption, a technique that allows computation on encrypted data, hinders an attacker's learning process and reduces its capabilities to attack the system. Specifically, we show that an attacker must solve challenging lattice problems to find attacks that are difficult to detect. Additionally, we show how the detection probability is affected by the attacker's solution to the problems and what parameters of the encryption scheme can be tweaked to increase the detection probability. We also develop a novel method that enables anomaly detection over homomorphically encrypted data without revealing the actual signals to the detector, thereby discouraging attackers from launching attacks on the detector. The detection can be performed using a hypothesis test. However, special care must be taken to ensure that fresh samples are used to detect changes from nominal behavior., Cyberattacker mot kritisk infrastruktur utgör ett allvarligt hot mot samhället, eftersom de kan få förödande konsekvenser för ekonomin, säkerheten eller folkhälsan. Dessa infrastrukturer utgörs ofta av ett stort nätverk av cyberkomponenter, såsom sensorer, styrenheter, datorer och kommunikationsenheter, för att övervaka och styra sina fysiska processer. En angripare kan utnyttja sårbarheterna i dessa cyberkomponenter för att få tillgång till systemet och därefter manipulera dess beteende eller funktionalitet. Denna avhandling behandlar och föreslår metoder som kan användas som en första försvarslinje mot sådana attacker för cyberfysiska system. I den första delen av avhandlingen undersöker vi hur oinformerade angripare kan lära sig att attackera ett cyberfysiskt system genom att avlyssna dem via cyberkomponenten. Genom att lära sig att manipulera det fysiska systemet kan angriparen ta reda på hur man kan förstöra, eller helt ta över det cyberfysiska systemet utan att något alarm går av förrän det är försent. Genom att stoppa angriparen i inlärningsfasen tvingas angriparen att agera mer omedvetet, vilket ökar chanserna att upptäcka dem. Vi analyserar hur homomorf kryptering, vilket är en krypteringsmetod som möjliggör beräkning med krypterad data, hindrar angriparens inlärnings-process och minskar dess förmåga att attackera systemet. Specifikt visar vi att en angripare måste lösa svåra gitterproblem för att hitta svårdetekterade cyberattacker. Dessutom visar vi hur detektionssannolikheten påverkas av hur bra angriparens lösning är och vilka parametrar i krypteringsschemat som kan justeras för att öka sannolikheten att upptäcka anfallet. Vi utvecklar också en ny metod som möjliggör anomalidetektering över homomorft krypterade data, utan att avslöja de faktiska signalerna för detektorn och därmed avskräcka angripare från att attackera detektorn. Detektionen kan utföras med hjälp av ett hypotestest. Dock måste man se till att färska prover används för att upptäcka förän, QC 20231020

A multiphase induction machine model using vector space decomposition provides insights into many space harmonics through decoupled reference frames. These decoupled reference frames host specific space vectors related to particular space harmonics. Based on the physical winding configuration, these vector spaces can be excited independently and simultaneously for the production of torque. However, this approach may result in beat oscillations due to interference between excited vector spaces if proper synchronization of vector spaces is not maintained. This paper describes this phenomenon through experimental tests. Furthermore, a solution eliminating the beat oscillations is proposed while optimizing the stator current or rotor flux linkage peaks. The effectiveness of the solution is experimentally verified on a 9-phase induction machine., QC 20231114

Autonomous Vehicle (AV) technology promises safer, greener, and more efficient means of transportation for everyone. AVs are expected to have their first big impact in closed environments, such as mining areas, ports, and construction sites, where Heavy-Duty Vehicles (HDVs) operate. This thesis addresses lateral motion control for autonomous HDVs using Model Predictive Control (MPC). Lateral control for HDVs still has many open questions to be addressed, in particular, precise path tracking while ensuring a smooth, comfortable, and stable ride, coping with both external and internal disturbances, and adapting to different vehicles and conditions. To address these challenges, a comprehensive control module architecture is designed to adapt seamlessly to different vehicle types and interface with various planning and localization modules. Furthermore, it is designed to address system delays, maintain certain error bounds, and respect actuation constraints. This thesis presents the Reference Aware MPC (RA-MPC) for autonomous vehicles. This controller is iteratively improved throughout the thesis. The RA-MPC introduces a method to systematically handle references generated by motion planners which can consider different algorithms and vehicle models from the controller. The controller uses the linear time-varying MPC framework and considers control input rate and acceleration constraints to account for steering limitations. Furthermore, multiple models and control inputs are considered throughout the thesis. Ultimately, curvature acceleration is used as the control input, which together with stability ingredients, allows for stability guarantees under certain conditions via Lyapunov techniques. MPC is highly dependent on the prediction model used. This thesis proposes and compares different models. First, an offline-fitted, vehicle-specific nonlinear curvature response function is proposed and integrated into the kinematic bicycle model. The curvature response function, Teknik för autonoma fordon lovar säkrare, grönare och effektivare transportmedel för alla. Autonoma fordon förväntas få sin första stora inverkan i inhägnade områden, såsom gruvområden, hamnar och byggarbetsplatser, där tunga fordon är i drift. Denna avhandling behandlar lateral rörelsereglering för autonoma tunga fordon med hjälp av modellprediktiv reglering (Model Predictive Control, MPC). Lateral rörelsereglering för tunga fordon har fortfarande flera öppna utmaningar, i synnerhet gällande precis banföljning som garanterar en mjuk, komfortabel och stabil resa, hanterar både externa och interna störningar och anpassar sig till olika fordon och förhållanden. För att möta dessa utmaningar är en omfattande regleringsarkitektur utformad för att sömlöst anpassa sig till olika fordonstyper och gränssnitt mot olika planering- och lokaliseringsmoduler. Därutöver är arkitekturen utformad för att hantera systemfördröjningar, bibehålla särskilda felmarginaler och respektera styrdons begränsningar. Denna avhandling presenterar den Referensmedvetna modellprediktiva regulatorn (Reference Aware MPC, RA-MPC) för autonoma fordon. Denna regulator är iterativt för-bättrad genom hela avhandlingen. Den referensmedvetna modellprediktiva regulatorn introducerar en metod för att systematiskt hantera referenssignaler genererade av rörelseplanerare som beaktar andra typer av algoritmer och fordonsmodeller än regulatorn. Regulatorn använder det linjärt tidsvarierande MPC-ramverket och beaktar begränsningar på styrsignalsförändringar och accelerationer för att ta hänsyn till begränsningar på styrningen. Vidare beaktas flertalet modeller och styrsignaler genom hela avhandlingen. I slutändan används kurvaturacceleration som styrsignal, vilket tillsammans med stabilitetsingredienser möjliggör stabilitetsgarantier under särskilda förhållanden via Lyapunov-tekniker. MPC är starkt beroende av den prediktionsmodell som används. Denna avhandling föreslår och jämför ett flertal olika modeller. Först, QC 20231003

In this thesis, we study the related problems of reinforcement learning and optimal adaptive control, specialized to specific classes of stochastic and structured dynamical systems. By stochastic, we mean systems that are unknown to the decision maker and evolve according to some probabilistic law. By structured, we mean that they are restricted in some known way, e.g., they belong to a specific model class or must obey a set of known constraints. The objective in both problems is the design of an optimal algorithm, i.e., one that maximizes a certain performance metric. Because of the stochasticity, the algorithm faces an exploration-exploitation dilemma, where it must balance collecting information from the system and leveraging existing information to choose the best action or input. This trade-off is best captured by the notion of regret, defined as the difference between the performance of the algorithm and an oracle which has full knowledge of the system. In the first part of the thesis, we investigate systems that can be modeled as Markov Decision Processes (MDPs) and derive general asymptotic and problem-specific regret lower bounds for ergodic and deterministic MDPs. We make these bounds explicit for MDPs that: i) are ergodic and unstructured, ii) have Lipschitz transitions and rewards, and iii) are deterministic and satisfy a decoupling property. Furthermore, we propose Directed Exploration Leaning (DEL), an algorithm that is valid for any ergodic MDP with any structure and whose regret upper bound matches the associated regret lower bounds, thus being truly optimal. For this algorithm, we present theoretical regret guarantees as well as a numerical demonstration that verifies its ability to exploit the underlying structure. In the second part, we study systems with uncertain linear dynamics and which are subject to additive disturbances as well as state and input constraints. We develop Self-Tuning Tube-based Model Predictive Control (STTMPC), an adaptive, QC 20231003

A multiphase machine control implementing vector space decomposition or harmonic plane decomposition for modeling purposes can reconfigure the stator's current distribution to generate magnetomotive force waves with different pole pair numbers simultaneously. The pole pair numbers are an integer multiple of the pole pairs generated by the physical winding configuration. This property can be used for reconfiguring the pole-pair number in real time or increasing the system's torque density. This paper studies the modeling and the impact of symmetrical and asymmetrical winding configurations on the pole-pair number reconfiguration. Furthermore, pole number transition is experimentally verified on a proposed winding configuration., Part of proceedings ISBN 979-8-3503-9996-7QC 20231114

Multiphase electrical machine are inherently truefault tolerant, which is a key difference as compared to conventionalthree-phase machines. The subgroup of variable phase-polemachines have the ability to change phase-pole configurationsonly through current control. Due to the higher availabilityand simpler construction, three-phase voltage source invertersmodules are often used for multiphase electrical machines.While many approaches exist for one single open-phase fault,this paper proposes an open-fault detection method capable todetect the failure of an entire three-phase module for variablephase-pole machines. It is an extension to the harmonic-planedecompositionfault detection analyzing the instantaneous spaceharmonic content of the current distribution. Simulation resultsdemonstrate the feasibility of the approach., Part of ISBN 979-8-3503-2077-0QC 20231012

The uncertainty in the prediction calculated using the delta method for an over-parameterized (parametric) black-box model is shown to be larger or equal to the uncertainty in the prediction of a canonical (minimal) model. Equality holds if the additional parameters of the overparameterized model do not add flexibility to the model. As a conclusion, for an overparameterized black-box model, the calculated uncertainty in the prediction by the delta method is not underestimated. The results are shown analytically and are validated in a simulation experiment where the relationship between the normalized traction force and the wheel slip of a car is modelled using e.g., a neural network.

Neural networks are ubiquitous components of Machine Learning (ML) algorithms. However, training them is challenging due to problems associated with exploding and vanishing loss-gradients. Gradient clipping is shown to effectively combat both the vanishing gradients and the exploding gradients problems. As the name suggests, gradients are clipped in order to prevent large updates. At the same time, very small neural network weights are updated using larger step-sizes. Although widely used in practice, there is very little theory surrounding clipping. In this paper, we analyze two popular gradient clipping techniques-the classic norm-based gradient clipping method and the adaptive gradient clipping technique. We prove that gradient clipping ensures numerical stability with very high probability. Further, clipping based stochastic gradient descent converges to a set of neural network weights that minimizes the average scaled training loss in a local sense. The averaging is with respect to the distribution that generated the training data. The scaling is a consequence of gradient clipping. We use tools from the theory of dynamical systems for the presented analysis.

A new class of iterated linearization-based nonlinear filters, dubbed dynamically iterated filters, is presented. Contrary to regular iterated filters such as the iterated extended Kalman filter (IEKF), iterated unscented Kalman filter (IUKF) and iterated posterior linearization filter (IPLF), dynamically iterated filters also take nonlinearities in the transition model into account. The general filtering algorithm is shown to essentially be a (locally over one time step) iterated Rauch-Tung-Striebel smoother. Three distinct versions of the dynamically iterated filters are especially investigated: analogues to the IEKF, IUKF and IPLF. The developed algorithms are evaluated on 25 different noise configurations of a tracking problem with a nonlinear transition model and linear measurement model, a scenario where conventional iterated filters are not useful. Even in this “simple” scenario, the dynamically iterated filters are shown to have superior root mean-squared error performance as compared with their respective baselines, the EKF and UKF. Particularly, even though the EKF diverges in 22 out of 25 configurations, the dynamically iterated EKF remains stable in 20 out of 25 scenarios, only diverging under high noise.

Markvärme används för att hålla bland annat gångvägar och parkeringshus fria från snö och is. Dessa system styrs oftast baserat på momentan data hämtat direkt från anläggningen. Dessa markvärmesystem förbrukar stora mängder energi och syftet med detta projekt var att baserat på framtida väderprognoser optimalt och individuellt styra driften för att minimera energiförbrukningen. Baserat på momentan marktemperatur och framtida väderprognoser går det att skapa en modell över markens framtida temperatur. Med hjälp av den framtida marktemperaturen samt tidigare och framtida nederbördsintensiteten går det att identifiera potentiellt farliga förhållanden som kommer att kräva markvärme. När detta behov är identifierat beräknas den tid det kommer ta att värma upp marken till en bestämd temperatur och smälta all nederbörd som befinner sig på marken. När all nederbörd har smält börjar processen att torka marken. Med en ytavrinningsfaktor på 0.9 antas det att 90% av all nederbörd kommer att rinna av från vägen. Baserat på marktemperaturen och diverse olika väderparametrar beräknas avdunstningstiden för de resterande 10% av nederbörd som fortfarande är kvar på marken. När all nederbörd har avdunstat och ingen ny nederbörd är väntad inom tre timmar så anses marken vara torr och systemet går ur drift. Systemet testades med data från februari 2023 och jämfördes med driften av markvärmen hos en av GateIBS anläggningar. Logiken för start av drift testades separat från logiken för stopp av drift. Det nya systemet skulle starta driften två gånger under februari med en nederbördsgräns satt till 3mm och 5mm och tre gånger med nederbördsgränsen satt till 1mm. Detta är en drastisk minskning jämfört med det faktiska markvärmesystemet som gick i drift totalt tio gånger under samma månad. Logiken för stopp av drift testades enbart då det faktiska markvärmesystemet gick i drift och under hela månaden så skulle det nya systemet sammanlagt stoppat driften 50 timmar tidigare vilken är en mi, Ground heating is used to keep walkways and parking lots free from snow and ice. These systems are usually controlled based on real-time data obtained directly from the facility. These ground heating systems consume large amounts of energy, and the purpose of this project was to optimally and individually control the operation based on future weather forecasts in order to minimize energy consumption. Based on the current ground temperature and future weather forecasts, it is possible to create a model of the future temperature of the ground. Using the future ground temperature, as well as past and future precipitation intensity, it is possible to identify potentially hazardous conditions that will require ground heating. Once this need is identified, the time it will take to heat the ground to a specific temperature and melt all the precipitation on the ground is calculated. After all the precipitation has melted, the process of drying the ground begins. With a surface runoff factor of 0.9, it is assumed that 90% of all precipitation will run off from the road. Based on the ground temperature and various weather parameters, the evaporation time for the remaining 10% of precipitation still on the ground is calculated. When all the precipitation has evaporated and no new precipitation is expected within three hours, the ground is considered dry, and the system stops operating. The system was tested using data from February 2023 and compared with the operation of the ground heating system at one of GateIBS facilities. The logic for starting the operation was tested separately from the logic for stopping the operation. The new system would initiate the operation twice in February with a precipitation threshold set at 3mm and 5mm, and three times with the threshold set at 1mm. This is a drastic reduction compared to the actual ground heating system, which started a total of ten times during the same month. The logic for stopping the operation was tested only when t

This report focuses on the steering of a Very Narrow Aisle (VNA) forklift, which undergoes frequent transitions between manual and wire guidance modes. Manual operation is employed outside narrow aisles, while wire guidance mode is utilized within them.VNA forklifts are commonly employed in environments where space optimization and productivity are of utmost importance. The forklift’s steering system operates hydraulically through the movement of two cylinders controlled by a proportional valve, which in turn is controlled by a current input. To implement the steering control, a Speedgoat target machine, a rapid control prototyping platform, is utilized. The control algorithm is developed in Simulink Real-Time and integrated with the forklift’s MCU through the Speedgoat target machine, connected via a CAN bus. The Speedgoat control strategy utilizes the distance from the wire (DFW) and heading angle(HA) to generate a pivot angle request. In contrast, the current control strategy aims to minimize DFW, HA, and the pivot angle using P controllers. For the Speedgoat control strategy, the pivot angle request is compared to the current pivot angle to produce a steering command. A PD controller is applied to the heading angle for rapid stabilization of steering changes, while a PI controller is used to ensure the actual pivot angle follows the desired pivot angle. To minimize the distance from the wire, a P controller with two different settings, depending on proximity to wire locking is employed. The control strategy also incorporates bumpless transfer techniques to ensure smooth transitions between manual and wire guidance modes by gradually adjusting the impact of wire guidance and manual steering. Anti-windup measures are taken to prevent integral wind-up effects, and various PID tuning methods are explored to determine the optimal controller parameters. A simulation model is developed to simulate the manual steering of the forklift. The manual steering implemented in

Robust detection, localization, and tracking of objects are essential for autonomous driving. Computer vision has largely driven development based on camera sensors in recent years, but 3D localization from images is still challenging. Sensors such as LiDAR or RADAR are used to compute depth; each having its own advantages and drawbacks. The main idea of the project is to be able to mix images from the camera and RADAR detections in order to estimate depths for the objects appearing in the images. Fusion strategies can be considered the solution to give a more detailed description of the environment by utilizing both the 3D localization capabilities of range sensors and the higher spatial resolution of image data. The idea is to fuse 3D detections from the RADAR onto the image plane, this requires a high level of synchronization of the sensors and projections of the RADAR data on the required image., Robust detektering, lokalisering och spårning av objekt är avgörande för autonom körning. Datorseende har till stor del drivit utvecklingen baserad på kamerasensorer de senaste åren, men 3D-lokalisering från bilder är fortfarande utmanande. Sensorer som LiDAR eller RADAR används för att beräkna djup; var och en har sina egna fördelar och nackdelar. Huvudtanken med projektet är att kunna blanda bilder från kameran och RADAR-detektioner för att uppskatta djup för de objekt som förekommer i bilderna. Fusionsstrategier kan anses vara lösningen för att ge en mer detaljerad beskrivning av miljön med både 3D-lokaliseringsförmågan hos avståndssensorer och den högre rumsliga upplösningen av bilddata. Tanken är att smälta samman 3D-detektioner från RADAR till bildplanet, detta kräver en hög nivå av synkronisering av sensorerna och projektioner av RADAR-data på den önskade bilden.

Dagens samhälle står inför många utmaningar där energieffektivisering är en stor del för att nå Sveriges energipolitiska mål. Det finns anledning att effektivisera och optimera energianvändningen i Sveriges papper- och massaindustrier då de står för en stor andel av nationell energiförbrukning. Papper- och massatillverkning genererar processavloppsvatten som kräver rening då det innehåller en hög halt organiska ämnen som kan orsaka miljöproblem vid utsläpp till närliggande recipient. För att resurs- och miljöeffektivisera krävs god förståelse och kunskap kring hur den aktuella reningsprocessen ser ut och vilka påverkansfaktorer som spelar störst roll då de varierar från bruk till bruk. I det här examensarbetet undersöks möjligheten att effektivisera en biologisk reningsprocess på Stora Enso Skoghalls bruk med hänsyn till elanvändning för luftning och tillsättning av närsalter. På grund av en ny installering av luftningsteknik så har bruket förbättrat förutsättningarna för syresättning och har efter detta ett ökat överskott på syre i deras luftade damm. Installationen öppnade för att kunna reglera lufttillförseln och även en möjlighet att studera närsaltsdoseringen då dessa faktorer har stor påverkan på funktionen av det biologiska reningsverket. En beräkningsmodell över reningsprocessen formades efter processdata från bruket med målsättning att göra en modell som stämde väl överens med verkligheten. Olika scenarion har därefter studerats för att jämföra vilka åtgärder som har störst inverkan på miljöpåverkan, driftkostnad och energianvändning. Förändring av antalet ytluftare i drift efter syrebehov i den luftade dammen genomfördes som ett scenario. Minskning av slamålder undersöktes med hänsyn till luftning och närsaltsbehov i den luftade dammen. Även en sänkning av syrenivån i MBBR:en (biofilmreaktor med rörlig bädd) undersöktes samt installation av den nya luftningstekniken Hyperclassic i halva volymen av den luftade dammen, dels i kombination, dels som separata, To achieve the goals set by Sweden’s energy policy, improving the efficiency of energy use in society and economy is an imperative. Sweden’s paper and pulp sector accounts for a significant proportion of national energy usage. Paper and pulp production generates wastewater from the production process that requires purification as it has a high content of organic substances that can cause environmental problems when released to nearby sinks. In order to improve resource and environmental efficiency, a good understanding and knowledge of the effectiveness of the current purification process, and the main environmental impact categories (which vary from mill to mill), is required. In this degree project, the possibility of optimizing a biological purification process at the Stora Enso Skoghall mill, with respect to electricity usage and the addition of nutrients, has been investigated. By installing a new aeration technology, the mill has facilitated an increase in the level of oxygen in its aerated lagoon. The installation made it possible to regulate the air supply and provided an opportunity to study the dosing of nutrients – two factors that have a major impact on the functioning of the biological treatment plant. A computational model of the purification process was developed, based on process data obtained from the mill. The aim was to build a model that synchronized well with reality. Different scenarios were compared on the basis of environmental impacts, energy use and operational expenses. The number of surface aerators was changed according to oxygen demand in the aerated pond, in one scenario. The consequence of the reduction of the Sludge Retention Time (SRT) on aeration and nutrient salt requirements in the aerated lagoon, was investigated. The effects of lowering the oxygen level in the MBBR (Moving Bed Biofilm Reactor), and the installation of the new Hyperclassic aeration technology in half the volume of the aerated lagoon, were studied – both in combi

In order to meet the increasing demand for sustainable transportation, trains need to run with tighter schedules, more departures and more trains in depot. Multiple trains in depot has been linked to low frequency instability at many locations around the world. The instability originates in the interaction between the infrastructure and trains, specifically the line converter module (LCM). The phenomenon has been studied during the last two decades. Today there are a number of methods used to analyse low frequency stability, although these methods are usually slow and cumbersome. In this project a modelling approach is proposed for a linear small-signal DQ-frame admittance model of the LCM. The model describes the relationship of the low frequency oscillations between the line voltage and line current. Similar linear models have been derived previously, although the design of the specific train modelled is in part significantly different from the design of the trains modelled previously. The main difference is in the DQ-transformation for the phase locked loop (PLL), which in this thesis is implemented with a DFT-based method, and the implementation of the current controller (CC) which in this thesis is implemented in the stationary reference frame. The LCM and infrastructure (modelled as an impedance) is then analysed as a feedback loop using dominant poles analysis, regularly used in control theory. The result from the proposed model was evaluated using methods that are common in the industry. Furthermore, as the LCM is a complex system and the impact of different components and control parameters is not fully understood, the proposed model is used to examine how changing different parameters affect the overall stability of the system. The correlation between the results from the proposed model and the methods used for evaluation implies that the modelling approach is accurate. The two main benefits of the proposed model is that the method is much faster than most, För att möta den ökande efterfrågan på hållbar transport, behöver tåg idag köras med stramare tidtabeller, fler avgångar och fler tåg i depå. Flera tåg i depå har kopplats till lågfrekvent instabilitet på många platser runt om i världen. Instabiliteten har sitt ursprung i interaktionen mellan infrastrukturen och tågen, särskilt nätströmriktaren (LCM). Fenomenet har studerats under de senaste två decennierna. Idag finns det ett antal metoder som används för att analysera lågfrekvent stabilitet, dock är dessa metoder vanligtvis är långsamma och besvärliga. Detta projekt beskriver en modelleringsmetod lågfrekvenssvängningar mellan linjespänningen och linjeströmmen. Liknande linjära modeller har härletts tidigare, skiljer sig utformningen av det specifika tåget modellerad i denna avhandling från de tåg som modellerats tidigare i vissa design val. Den största skillnaden ligger i DQ-transformationen för den synkroniseringsregulatorn (PLL), som i denna avhandling är implementerad med en DFT-baserad metod, och implementeringen av strömregulatorn (CC) som i denna uppsats implementeras i den stationära referensramen. LCM tillsammans med infrastruktur (modellerad som en impedans) analyseras sedan som en stängd återkoppling där det stängda systemets poler härleds. Resultatet från den föreslagna modellen utvärderades med metoder som är vanliga i branschen. Dessutom, eftersom LCM är ett komplext system och effekten av olika komponenter och styrparametrar inte är helt förstod, används modellen för att undersöka hur förändringar i olika parametrar påverkar hela systemets stabilitet. Korrelationen mellan resultaten från det föreslagna modell och de metoder som används för utvärdering av lågfrekvent stabilitet idag innebär att modelleringsmetoden är lovande. De metoden är mycket snabbare än de flesta andra metoder för lågfrekvent analys och att den föreslagna modellen är användbar för att testa olika parameterkonfigurationer och förändringar i utformningen av LCM.

Hybrid stepper motors are small electrical motors with high torque production, compared with other electrical motors of the same size. Hybrid stepper motors are reliable in open-loop systems, in Sinusoidal mode, but with a drawback of high power consumption. The power consumption may be reduced by Field-Oriented Control, but this control mode requires a positioning sensor, adding size and cost to the system. This Master’s Thesis explores the possibilities of observer-based Field-Oriented Control on a two-phase hybrid stepper motor without using a positioning sensor, run on a microprocessor and executed during the interrupt scheduling routine, coded in C/C++.

In wastewater treatment plants, the amount of ammonium is one example of a measure to determine the quality of the effluent wastewater. Ammonium is regarded as a hazardous chemical for aqueous ecosystems and can cause eutrophication due to its high nitrogen content. The ammonium content in the treated wastewater is controlled by aeration of the biological treatment stage, in which ammonium is converted to nitrate. The aeration process often accounts for the largest energy consumption of the wastewater treatment plant, which motivates automatic control solutions that can both aid in reducing the discharge of ammonium in the effluent and improve the energy efficiency of the aeration process. One such control technique currently used by several large municipal wastewater treatment plants in Sweden is ammonium-based aeration control. In this technique, the aeration process is controlled based on measurements of the effluent ammonium concentration. The purpose of the thesis was to study an extension of ammonium-based aeration control that could better adapt to daily, and often large, fluctuations in the influent load. The proposed method is to use an iterative algorithm to tune the set-point of the ammonium feedback controller. The objective is to, over a given time interval, achieve a flow-proportional mean of the effluent ammonium concentration close to a desired value for a wide range of influent loads. The method was tested by extensive simulations, and the results indicate that the iterative set-point tuning algorithm has the potential to offer a superior ability to achieve a desired flow-proportional mean at the end of a given evaluation period and, in some instances, energy savings compared to standard ammonium feedback control.

The present paper discusses a new method ofclassifying kinds of material and driving stages and styles ofVolvo wheel loaders (WLO). This is achieved by indirectly monitoringrelevant, but usually latent variables, based on directlymonitored sensors of WLOs. The continuous classifications willsupport Volvo’s actual objectives such as, e.g., maximizingthe remaining useful life of components, fuel efficiency, andproductivity.To this end, a set of WLO machines was equipped with extrasensors and collected a limited dataset with richer information.Based on this limited dataset, different machine learning (ML)methods were tested to derive and to verify the classifications. Itshowed that support vector machines (SVM) produced the bestresults: the driving styles could be classified with a test accuracyof 77% (resp, 99.5%) in the loading (resp. unloading) drivingstage. Further the SVM model is also verified both theoreticallyto enhance the confidence in the model and experimentally witha set of additional test drivers.

Inom reglerteknik har integrationen av maskininlärningsmetoder framträtt som en central strategi för att förbättra prestanda och adaptivitet hos styrsystem. Betydande framsteg har gjorts inom flera viktiga aspekter av reglerkretsen, såsom inlärningsbaserade metoder för systemidentifiering och parameterskattning, filtrering och brusreducering samt reglersyntes. Denna avhandling fördjupar sig i området inlärning för reglerteknik med särskild betoning på inlärningsbaserade regulatorer och identifieringsmetoder. Avhandlingens första del behandlar undersökningen av neuronnätsbaserad Modellprediktiv Reglering (MPC). Olika nätstrukturer studeras, både generella black box-nät och nät som väver in MPC-specifik information i sin struktur. Dessa nät jämförs och utvärderas med avseende på två prestandamått genom experiment på realistiska två- och fyrdimensionella system. Den huvudsakliga nyskapande aspekten är inkluderingen av gradientdata i träningsprocessen, vilket visar sig förbättra noggrannheten av de genererade styrsignalerna. Vidare påvisar de experimentella resultaten att en MPC-informerad nätstruktur leder till förbättrad prestanda när mängden träningsdata är begränsad. Med insikt om vikten av noggranna matematiska modeller av styrsystemet, riktar den andra delen av avhandlingen sitt fokus mot inlärningsbaserade identifieringsmetoder. Denna forskningsgren behandlar karakterisering och modellering av dynamiska system med hjälp av maskininlärning. Avhandlingen bidrar till området genom att introducera kooperativa systemidentifieringsmetoder för att förbättra parameterskattningen. Specifikt utnyttjas verktyg från Optimal Transport för att introducera en ny och mer generell formulering av ramverket Correctional Learning. Detta ramverk är baserat på en mästare-lärlingsmodell, där en expertagent (mästare) observerar och modifierar den insamlade data som används av en lärande agent (lärling), med syftet att förbättra lärlingens skattningsprocess. Genom att formulera correct, In the context of control systems, the integration of machine learning mechanisms has emerged as a key approach for improving performance and adaptability. Notable progress has been made across several aspects of the control loop, including learning-based techniques for system identification and estimation, filtering and denoising, and controller design. This thesis delves into the rapidly expanding domain of learning in control, with a particular focus placed on learning-based controllers and learning-based identification methods. The first part of this thesis is devoted to the investigation of Neural Network approximations of Model Predictive Control (MPC). Model-agnostic neural network structures are compared to networks employing MPC-specific information, and evaluated in terms of two performance metrics. The main novel aspect lies in the incorporation of gradient data in the training process, which is shown to enhance the accuracy of the network generated control inputs. Furthermore, experimental results reveal that MPC-informed networks outperform the agnostic counterparts in scenarios when training data is limited. In acknowledgement of the crucial role accurate system models play in in the control loop, the second part of this thesis lends its focus to learning-based identification methods. This line of work addresses the important task of characterizing and modeling dynamical systems, by introducing cooperative system identification techniques to enhance estimation performance. Specifically, it presents a novel and generalized formulation of the Correctional Learning framework, leveraging tools from Optimal Transport. The correctional learning framework centers around a teacher-student model, where an expert agent (teacher) modifies the sampled data used by the learner agent (student), to improve the student's estimation process. By formulating correctional learning as an optimal transport problem, a more adaptable framework is achieved, better suited for e, VR 2018-03438 projekt 3224