Energy Digest

A two-stage Bayesian Neural Network (BNN) surrogate is introduced to solve the AC optimal power flow problem with uncertainty, enabling robust performance under novel operating conditions. The model integrates an on-the-fly learning mechanism that dynamically retrains when prediction accuracy degrades, ensuring strong generalization capability across different network sizes and stochastic renewable injections. The framework offers practical value as a real-time advisory tool for system operators and fast initializer for conventional solvers.

Distributed economic dispatch schemes for isolated battery energy storage systems (BESSs) are developed to alleviate capacity degradation and power loss. The proposed schemes utilize proportional-integral protocols with reset mechanisms (PI+R protocols), which improve consensus rate and control accuracy. Simulation results show that the second scheme outperforms the first in terms of consensus rate, convergence rate, and overall system performance.

A high-fidelity load modeling framework is used to disaggregate facility demand into IT and non-IT components and represent load behavior at second- and minute-level resolutions. A comprehensive optimization framework is developed to plan and design battery energy storage systems (BESS) for behind-the-meter data center microgrids, reducing operational stress by smoothing load spikes and excursions. The proposed methodology co-optimizes BESS with various power generation assets while incorporating operational constraints, resulting in cost-effective, resilient, and operationally feasible designs.