Efficient control of HVAC (Heating, Ventilation, and Air Conditioning) systems is crucial for balancing demand and supply of energy in buildings, particularly during peak demand pe-riods. This dissertation aims to address three research gaps. First, previous research effortshave focused on decreasing energy consumption over peak time while considering comfort asa fixed range of temperatures or using generic indices for a population rather than focusingon individual thermal preferences. In response to this gap, a novel occupant-centric con-trol strategy is proposed to minimize energy costs while prioritizing personalized comfort.The proposed controller is tested in a simulation environment under different contextualconditions and in a real-world testbed. Second, another challenge of the existing HVACsystem controllers is finding the right balance between energy cost and occupant comfort inco-optimization formulations. The proposed balance should be adapted to different environ-ments. To address this challenge, an evolutionary Reinforcement Learning (RL) approachis introduced that enables the system to learn and adapt the trade-off coefficient betweenenergy and comfort optimization, enhancing the system's adaptability to different environ-mental and contextual conditions. Third, existing load flexibility models mainly considerspace-related factors and often overlook individual preferences. In the final phase, we shiftour focus from spaces to people and examine how current load flexibility models may affectindividual thermal comfort. Also, we devise a feature to predict load-shedding potentialbased on user properties. The performance of these three frameworks/models is assessedthrough a comprehensive uncertainty quantification analysis, taking into account the di-versity in occupants' preferences and the number of individuals present. Furthermore, theproposed approaches are compared with benchmark controllers from existing literature in asimulated environment. To validate their feasibility, a real-world experiment in an apart-ment unit as a practical test-bed is conducted. This research aims to improve the energyefficiency of HVAC systems, improve overall comfort experience, and evaluate the effect ofindividual comfort based on the current load flexibility models. / Doctor of Philosophy / Heating, Ventilation, and Air Conditioning (HVAC) systems are essential for maintaining comfort in buildings but account for a significant portion of energy use, especially during peak demand times when energy consumption is at its highest. Optimizing HVAC systems can help reduce costs and energy peaks, yet traditional approaches often overlook the diverse comfort needs of occupants. As buildings become more connected to the energy grid, there is a growing need for smarter, occupant-centered HVAC strategies that balance energy savings with personalized comfort. This research explores innovative solutions to address these chal-lenges. First, it introduces advanced control strategies that incorporate individual comfortpreferences into HVAC systems. These strategies help reduce energy peaks while ensuringoccupants remain comfortable, creating systems that are more efficient and user-friendly.Second, the study develops adaptable technologies, using cutting-edge artificial intelligence,to make HVAC systems smarter. These systems can learn and adjust to changing environ-ments, making them more effective in both simulations and real-world scenarios. Finally,the research shifts focus from buildings as a whole to the people inside them. It evaluateshow energy-saving programs affect individual comfort and proposes new tools to predict andoptimize energy savings without sacrificing comfort. By combining personalized comfortmodels, adaptable technologies, and user-focused strategies, this work paves the way forbuilding systems that save energy, reduce costs, and prioritize the well-being of occupants.The findings not only highlight practical solutions for today's energy challenges but also offera glimpse into the future of sustainable and adaptive building design.

Since the last United Nations Climate Change Conference in 2015 in Paris (the COP 21), world leaders acknowledged climate change. There is no need any more to justify the switch from fossil fuel-based to renewable energy sources. Nevertheless, this transition is far from being straightforward. Besides technologies that are not yet mature -- or at least not always financially viable in today's economy -- the power grid is currently not ready for a rapid and massive integration of renewable energy sources. A main challenge for the power grid is the inadequacy between electric production and consumption that will rise along with the integration of such sources. Indeed, due to their dependence on weather, renewable energy sources are intermittent and difficult to forecast with today's tools. As a commodity, electricity is a quite distinct good for which there must be perfect adequacy of production and consumption at all time and characterized by a very inelastic demand. High shares of renewable energy sources lead to high price volatility and a higher risk to jeopardize the security of supply. Additionally, the switch to renewable energy sources will lead to an electrification of loads and transportation, and thus the emergence of new higher-consumption loads such as electric vehicles and heat pumps. These new and higher-consumption loads, combined with the population growth, will cause over-rated power load increases with less predictable load patterns in the future.This work focuses on issues specific to the distribution power grid in the context of the current energy transition. Traditional low-voltage grids are perhaps the most passive circuits in power grids. Indeed, they are designed primarily using a fit and forget approach where power flows go from the distribution transformer to the consumers and no element has to be operated or regularly managed. In fact, low-voltage networks completely lack observability due to very low monitoring. The distribution grid will especially undergo drastic changes from this energy transition. Distributed sources and new high-consumption -- and uncoordinated -- loads result in new power flow patterns, as well as exacerbated evening peaks for which it is not designed. The consequences are power overloads and voltage imbalances that deteriorate grid components, such as a main asset like the medium-to-low voltage transformer. Additionally, the distribution grid is characterized by end-users that pay a price for electricity that does not reflect the grid situation -- that is, mostly constant over a year -- and allow little to no actions on their consumption.These issues have motivated authorities to propose a global approach to ensure security of electricity supply at short and medium-term. The latter requires, among others, the development of demand response programs that encourage users to take advantage of load flexibility. First, we propose adequate electricity pricing structures that will allow users to unlock the potential of such demand response programs; namely, dynamic pricings combined with a prosumer structure. Second, we propose a fast and robust two-level optimization, formulated as a mixed-integer linear program, that coordinates flexible loads. We focus on two types of loads; electric vehicles and heat pumps, in an environment with solar PV panels. The lower level aims at minimizing individual electricity bills while, at the second level, we optimize the power load curve, either to maximize self-consumption, or to smoothen the total power load of the transformer. We propose a parametric study on the trade-off between only minimizing the individual bills versus only optimizing power load curves, which have proven to be antagonist objectives. Additionally, we assess the impact of the rising share of flexible loads and renewable energy sources for scenarios from today until 2050. A macro-analysis of the results allows us to assess the benefits of load flexibility for every actor of the distribution grid, and depending on the choice of a pricing structure. Our optimization has proved to prevent evening peaks, which increases the lifetime of the distribution transformer by up to 200%, while individual earnings up to 25% can be made using adequate pricings. Consequently, the optimization significantly increases the power demand elasticity and increases the overall welfare by 10%, allowing the high shares of renewable energy sources that are foreseen. / Doctorat en Sciences de l'ingénieur et technologie / info:eu-repo/semantics/nonPublished

Im industriellen Bereich erfolgt die Energiebereitstellung auf thermischer, wie elektrischer Seite zunehmend mittels hocheffizienter Kraft-Wärme-Koppelung (KWK). Konventionelle KWK-Anlagen ohne Dampfturbine (DT) verfügen technologiebedingt über eine relativ starre, lastabhängige Stromkennzahl. Damit kann eine wärme- und/oder stromseitige Volatilität schlecht kompensiert werden ohne die jeweils gekoppelte Größe zu beeinflussen. Der integrierte Gas-Dampf-Prozess (GiD-Prozess) zeichnet sich aufgrund der halboffenen Prozessgestaltung durch eine anlagentechnisch sehr einfache Bauweise und damit gegenüber einer klassischen Gas und Dampf-Prozess (GuD)-Anlage geringeren Investitions- und Wartungskosten aus. Die vorliegende Arbeit befasst sich mit der Analyse des lastabhängigen Betriebsverhaltens des integrierter Gas-Dampf-Prozess (GiD)-Prozesses unter Berücksichtigung der Teillastfähigkeit und erreichbarer Lastgradienten. Hierzu werden umfangreiche Versuchsfahrten des Versuchskraftwerks am Zentrum für Energietechnik (ZET) der TU Dresden dargelegt und analysiert. Die Versuche werden durch transiente Systemsimulationen auf Komponentenebene der Kraftwerksanlage nachvollzogen und Erkenntnisse zu dem Effekt der Lastgradienten auf heißgasbeaufschlagte Bauteile gewonnen.