Data-Centric Power System Analysis with Generative Models

Modern power systems increasingly rely on data-driven tools for load analysis, monitoring, and operational decision-making. However, utilities and system operators often face practical challenges such as limited data availability, missing measurements, privacy constraints, and difficulty capturing rare but critical events. These issues directly limit the effectiveness of advanced analytics in real-world deployments. This talk presents a series of generative and data-driven methods developed to address these challenges using real utility data. First, MultiLoad-GAN is introduced to generate realistic synthetic load profiles that preserve temporal behavior and customer-level diversity, enabling large-scale studies without exposing sensitive data. Second, BERT-PIN and related language-model-based approaches are demonstrated for restoring missing smart meter data, improving data quality for downstream tasks such as load forecasting and planning. Third, the application of pre-trained large language models (LLMs) is discussed, showing how they can be adapted for power system time-series problems with reduced data and training requirements. Finally, a VAE–GAN framework is presented for event detection and classification using synchrophasor data, enabling early identification of abnormal system behavior.
This presentation will count for 1 Professional Development Hour (PDH) for the PE License in Wisconsin and Michigan.
Speaker(s): Dr. Yi Hu ,
Agenda:
5:00 Featured Speaker – Dr. Yi Hu
D.J. Bordini Center at FVTC
6:30 pm CST Social-Happy Hour at Cheddar's Scratch Kitchen
4531 W Wisconsin Ave
Appleton, WI
6:45 pm CST Dinner at Cheddar's Scratch Kitchen
Cost: $20, payable at registration
7:00 pm CST Short business meeting
Door prize drawing
Room: BC112A, Bldg: D.J. Bordini Center at FVTC, 5 N. Systems Drive, Appleton, Wisconsin, United States, 54914