Modeling and Control of Thermo-Electrical Microgrids Considering Uncertainties

Abstract

Global decarbonization targets for 2050 have accelerated the development of low-carbon energy system solutions. Consequently, numerous initiatives have been proposed to reduce carbon emissions, such as the deployment of high-efficiency heating and cooling systems based on Heat Pumps (HPs) and latent Thermal Energy Storage Systems (TESSs); the electrification of transportation systems, including Electric Vehicles (EVs) and Electric Aircraft (EA) operations in airports; and the integration of Renewable Energy Sources (RESs). To support the integration of these technologies, Microgrids (MGs) have emerged as a key architectural solution for coordinating renewable generation, electrical and thermal resources, and loads while delivering technical, economic, and environmental benefits.

This thesis develops detailed models to represent the Thermo-Electrical (TE) operation of building-integrated MGs, with a focus on residential and airport hangar applications, considering uncertainties and multi-zone building thermal dynamics with their associated thermodynamic and physical properties. Based on these models, Energy Management System (EMS) formulations are proposed to optimize the coordinated operation of electrical and thermal resources under practical operational constraints.

The first part of the thesis develops and validates an optimization-based EMS for a residential TE-MG that integrates an enthalpy-based model of a Phase-Change Material (PCM) TESS capable of operating in both active and passive modes. The proposed framework is formulated to minimize operating costs while maintaining indoor thermal comfort, with uncertainties in demand and environmental conditions addressed through a Model Predictive Control (MPC) approach and with explicit consideration of battery degradation. The EMS is applied to a real-world residential MG corresponding to the Energy Smart Home Lab (ESHL) at the Karlsruhe Institute of Technology (KIT) in Germany. Simulation results demonstrate the effective integration of the PCM system within the TE-MG operation and highlight its contribution to cost-effective and reliable energy management under various environmental conditions.

The second part of this thesis discusses the modeling of an airport hangar MG and an optimization-based EMS to coordinate the dispatch of the MG's TE resources, using an MPC approach to address uncertainties, and including a detailed building thermal model, HPs for heating and cooling, and battery degradation. The proposed mathematical model of the EMS is applied to a detailed model of an actual MG under development at the Waterloo Wellington Flight Centre (WWFC) in Ontario, Canada. The presented results demonstrate that the proposed framework enables reliable and cost-effective operation while ensuring multi-room thermal comfort, and achieves significant reductions in operational costs and CO2 emissions compared to a baseline scenario without a MG and to a MG configuration employing simplified single-zone thermal modeling.

As the energy management of TE-MGs becomes increasingly challenging for model-based approaches due to detailed component modeling requirements and uncertainty in renewable generation and environmental conditions, the final part of this thesis proposes a model-free Reinforcement Learning (RL) framework, based on Deep Reinforcement Learning (DRL) methods, for the operation of multi-zone airport hangar MGs. Constraint satisfaction is ensured through the incorporation of physics-based dynamic constraints and a discretization of the coupled multi-zone thermal dynamics, which enables stable representation of inter-zone heat exchanges. Simulation results based on the operation of an actual Canadian airport MG are benchmarked against the proposed optimization-based approach, demonstrating that the physics-based RL framework achieves near-optimal performance. Furthermore, compared to conventional reward-based RL approaches, the proposed framework is shown to yield significantly faster convergence and more stable training behavior.