Aggregation of Thermostatically Controlled Loads for Fast Power System Services: From Theory to Practice

In electric power systems, mismatch between scheduled generation and demand leads to deviations from the nominal grid frequency, thereby posing risks to the reliability of the electric power system. The significant uncertainty introduced to the grid due to the intermittent nature of renewable energy resources, such as wind and solar, can exacerbate these frequency deviations. To mitigate that, additional flexible resources, such as load aggregations, are required to maintain the reliability of the system. Thermostatically Controlled Loads (TCLs) are excellent candidates for providing flexibility to the grid due to their thermal inertia and ubiquity across the distribution system. This dissertation addresses some of the main challenges associated with large-scale control of TCLs for providing fast power system services, i.e., on the order of seconds. These challenges emanate from a variety of sources, including the nonlinear nature of the underlying devices as well as the technical limitations that currently exist in practical systems. To that end, this dissertation develops advanced modelling, estimation, and control approaches tailored to address these challenges. With the goal of establishing more credibility for this technology, real-world experiments were carried out to better understand the limitations of the currently developed approaches and identify the main impediments to widespread adoption of load control. More specifically, this dissertation shows that TCLs can be successfully controlled to provide frequency regulation services with minimal impact on the end user. Since current state-of-the-art approaches do not consider some important aspects of the underlying devices, such as the presence of multiple zones and actuation delays, a range of aggregate models and control approaches are developed to address these deficiencies. Moreover, because technical limitations may prevent devices from being engaged very frequently, a computationally tractable optimization-based approach is developed for maximizing the potential of TCLs for providing fast power system services under infrequent actuation. This approach is shown to satisfy performance requirements imposed by system operators even with very infrequent actuation. To gain a deeper understanding of controller performance limits and potential undesired phenomena that can arise across different regimes during aggregate TCL control, this dissertation carries out a frequency response analysis of a probabilistic control scheme. It is shown that rapid switching commands can induce oscillations in the power output due to the inherent protective mechanism of the underlying devices. It is also demonstrated that highly detailed aggregate models are required to capture this behavior and enable deeper understanding of the control boundaries to avoid the introduction of undesirable effects on the grid. To reliably estimate the thermal parameters of individual TCL models, an identifiability analysis is carried out to more thoroughly explore the underlying models. This dissertation shows that commonly used individual TCL models are not identifiable, and subset selection is required to find an identifiable subset. The importance of identifiability is highlighted using a biased initialization scheme. Finally, a novel nonlinear least-squares problem is formulated and solved to estimate the set of identifiable parameters. More broadly, this dissertation attempts to establish more credibility for large-scale coordination of TCLs at fast timescales. It develops approaches to address some of the main challenges of TCL control by considering several practical aspects identified through real-world experiments. No major technological impediments are found that currently prevent TCLs from providing fast power system services while meeting performance requirements.