Exploiting Building Demand Flexibility Through Machine Learning for Building-to-grid Integration
Date of Award
Doctor of Philosophy (PhD)
Mechanical and Aerospace Engineering
Building-to-grid integration, Demand flexibility, Machine learning
Engineering | Mechanical Engineering
Demand flexibility – the ability to adjust a building's load profile across different timescales – is a key aspect of the ongoing effort to increase interconnectivity between buildings and the power grid. By harnessing their demand flexibility, buildings can provide significant benefits to the grid and bolster grid resilience and reliability. To facilitate the transition toward the "smart grid", new and intelligent control approaches are required that can seamlessly integrate building, occupant, and grid data and effectively control multiple building assets to provide grid services while maintaining occupants' required thermal comfort levels and reducing the building's overall energy consumption and costs. Current state-of-the-art control methods such as model predictive control have been shown to be effective for achieving these objectives; however, conventional control approaches use physics-based building models which can prove costly and time-consuming to develop and implement. Therefore, in order to facilitate the transition to the smart grid, building controls must be implemented in a way that is smart, flexible, fast, and scalable. This thesis seeks to address these issues by introducing several data-driven approaches to grid-interactive building control. The aim of the research is to improve building controls by developing and validating machine learning-based controllers that can fulfill the requirements listed above. A large-scale simulation framework for building-to-grid integration is introduced. By integrating buildings, distributed energy resources, and the power distribution network and co-optimizing over all components, the framework allows for simultaneous satisfaction of building and grid objectives while maintaining occupant thermal comfort. The framework is validated in several large-scale simulation experiments. It is also shown to be robust to uncertainties in certain uncontrollable variables. A purely data-driven, deep learning-based building model is introduced for control applications. The data-driven model is easy to train, given sufficient historical data. It is also easily embedded in a model predictive control framework, and simulation studies indicate it effectively leverages demand flexibility in a way that can provide grid services. A small-scale field experiment in a unique lab facility further validates the simulation results and demonstrates the value of the data-driven controller. This work contributes to the understanding and advancement of advanced building controls by addressing research gaps that have been identified in the literature. It may serve as a starting point from which more research may be done which incorporates black-box building models into large-scale simulations and field implementations for providing grid services through building demand flexibility. The work presented in this thesis serves to advance the grid-interactive building (GEB) initiative and will help to facilitate and accelerate the transition from the current building-grid paradigm to the smart grid.
Fontenot, Hannah Charlene, "Exploiting Building Demand Flexibility Through Machine Learning for Building-to-grid Integration" (2021). Dissertations - ALL. 1361.
Available for download on Wednesday, November 01, 2023