Congratulations to Dr. Thiha Aung for completing his PhD

Title: Applications of Gaussian Processes to Stochastic Control and Optimization for Grid-Scale Battery Energy Storage Systems

Abstract: Grid-scale battery energy storage systems require decision-making under uncertainty while respecting hard operational constraints. We develop Gaussian-process methods for two related stochastic control and optimization problems in grid-scale battery operation. The first contribution is SHADOw-GP, a backward actor–critic method that learns continuation-value and control emulators from simulated transitions. It accommodates continuous states and actions, nonlinear objectives, non-Gaussian dynamics, and hard constraints, and is applied to two complementary battery applications: renewable firming and energy arbitrage in electricity markets. The second contribution is ARBO-DART, a Bayesian optimization framework for battery co-optimization across day-ahead and real-time electricity markets. In this setting, evaluating each day-ahead dispatch profile requires SHADOw-GP to solve an expensive stochastic control problem for recourse in the real-time electricity market. ARBO-DART exploits the block structure of battery dispatch and adaptively refines the day-ahead parameterization around influential operating intervals. Numerical experiments demonstrate the efficacy of SHADOw-GP through comparisons with benchmark methods and battery-dispatch studies using synthetic test cases and empirical data, while ARBO-DART identifies high-value DART dispatch profiles using a compact, adaptively refined search space.

Committee Members: Mike Ludkovski (Chair), Tomoyuki Ichiba, Ruimeng Hu