Research in building agents by employing Large Language Models (LLMs) for computer control is expanding, aiming to create agents that can efficiently automate complex or repetitive computational tasks. Prior works showcased the potential of Large Language Models (LLMs) with in-context learning (ICL). However, they suffered from limited context length and poor generalization of the underlying models, which led to poor performance in long-horizon tasks, handling multiple applications and working across multiple domains. While initial work focused on extending the coding capabilities of LLMs to work with APIs to accomplish tasks, a new body of work focused on Graphical User Interface (GUI) manipulation has shown strong success in web and mobile application automation. In this work, I introduce LUCI: Large Language Model-assisted User Control Interface, a hierarchical, modular, and efficient framework to extend the capabilities of LLMs to automate GUIs. LUCI utilizes the reasoning capabilities of LLMs to decompose tasks into sub-tasks and recursively solve them. A key innovation is the application-centric approach which creates sub-tasks by first selecting the applications needed to solve the prompt. The GUI application is decomposed into a novel compressed Information-Action-Field (IAF) representation based on the underlying syntax tree. Furthermore, LUCI follows a modular structure allowing it to be extended to new platforms without any additional training as the underlying reasoning works on my IAF representations. These innovations alongside the `ensemble of LLMs' structure allow LUCI to outperform previous supervised learning (SL), reinforcement learning (RL), and LLM approaches on Miniwob++, overcoming challenges such as limited context length, exemplar memory requirements, and human intervention for task adaptability. LUCI shows a 20% improvement over the state-of-the-art (SOTA) in GUI automation on the Mind2Web benchmark. When tested in a realistic setting with over 22 commonly used applications, LUCI achieves an 80% success rate in undertaking tasks that use a subset of these applications. I also note an over 70% success rate on unseen applications, which is a less than 5% drop as compared to the fine-tuned applications.

As the share of variable renewable energy generation in the power system increases, there is a growing need for flexible resources to balance the resulting variability. Although many systems are transitioning away from fossil fuels, open-cycle gas turbines are likely to fill this balancing role for some time. Accordingly, accurate production cost modeling of the operational parameters of gas turbines will be increasingly crucial as these units are relied on more heavily for flexibility. This thesis explores the impact of three additional parameters—start-up profiles/costs, run-up rates, and forced outage rates—in the production cost modeling of a system as it adopts higher levels of wind and solar. Using PLEXOS simulations of the publicly available National Renewable Energy Laboratory’s 118 bus test system, the study examines how higher the increase in parameter modeling affects outcomes such as the number of start-ups and shut-downs, ramping, total generation costs for open-cycle gas turbines, and system-wide costs in three variable renewable energy penetration scenarios. The outcome of replacing certain conventional generation units with newer and more flexible combustion turbines is also examined. The results suggest the importance of detailed parameter modeling and continued research on the formulation of production cost models for flexible generation resources such as combustion turbines.

There has been a significant growth in the distributed energy resources (DERs) connected to the distribution networks in recent years. For a distribution system with a high penetration of DERs, a detailed modeling and representation of the distribution network is needed to accurately assess its steady-state and dynamic behavior. In this dissertation, a field-validated model for a real sub-transmission and distribution network is developed, including one of the feeders modeled with the secondary network and loads and solar PV units at their household/user location. A procedure is developed combining data from various sources such as the utility database, geoinformation data, and field measurements to create an accurate network model. Applying a single line to ground fault to the detailed distribution feeder model, a high voltage swell, with potentially detrimental impacts on connected equipment, is shown in one of the non-faulted phases of the feeder. The reason for this voltage swell is analyzed in detail. It is found that with appropriate control the solar PV units on the feeder can reduce the severity of the voltage swell, but not entirely eliminate it. For integrated studies of the transmission-distribution (T&D) network, a T&D co-simulation framework is developed, which is capable of power flow as well as dynamic simulations, and supports unbalanced modeling and disturbances in the distribution as well as the sub-transmission system. The power flow co-simulation framework is developed as a module that can be run on a cloud-based platform. Using the developed framework, the T&D system response is studied for balanced and unbalanced faults on the distribution and sub-transmission system. For some faults the resultant loss of generation can potentially lead to the entire feeder tripping due to high unbalance at the substation. However, it is found that advanced inverter controls may improve the response of the distribution feeders to the faults. The dissertation also highlights the importance of modeling the secondary network for both steady-state and dynamic studies. Lastly, a preliminary investigation is conducted to include different dynamic elements such as grid-forming inverters in a T&D network simulation.