Comparing Machine-Learning Algorithms used to Optimize Power Systems.

Presenter's Name(s)

Jack Colby
Ellison Fortin

Conference Year

2024

Abstract

As the integration of renewable energy resources to large-scale power systems rises, operating these systems has become increasingly challenging. Power system operation requires robust numerical solution algorithms to solve AC power flow equations. Machine-learning (ML) training algorithms, such as Adam (an ML-based numerical solver used to train GPT-3), have been implemented to solve such problems. Our research focuses on two innovations in ML training algorithms as prospective solutions: adaptive step-size tuning applied to an Adam-based solution algorithm, and the application of other ML algorithms to solve power flow. We compare results on several large-scale power system test cases.

Primary Faculty Mentor Name

Sam Chevalier

Status

Undergraduate

Student College

College of Engineering and Mathematical Sciences

Program/Major

Electrical Engineering

Primary Research Category

Engineering and Math Science

Abstract only.

Share

COinS
 

Comparing Machine-Learning Algorithms used to Optimize Power Systems.

As the integration of renewable energy resources to large-scale power systems rises, operating these systems has become increasingly challenging. Power system operation requires robust numerical solution algorithms to solve AC power flow equations. Machine-learning (ML) training algorithms, such as Adam (an ML-based numerical solver used to train GPT-3), have been implemented to solve such problems. Our research focuses on two innovations in ML training algorithms as prospective solutions: adaptive step-size tuning applied to an Adam-based solution algorithm, and the application of other ML algorithms to solve power flow. We compare results on several large-scale power system test cases.