Automatic generation control based on probably approximately correct Bayesian soft actor-critic algorithm with standard Q target

doi:10.1016/j.epsr.2025.112578

Automatic generation control based on probably approximately correct Bayesian soft actor-critic algorithm with standard Q target

2025-12-10

0

PRE

AI

DOI：10.1016/j.epsr.2025.112578

Original

Original request for help

Save

Abstract

En 中文

• A novel hybrid algorithm is proposed, integrating PAC-Bayesian theory for robust uncertainty-aware learning with a Standard Q-Target strategy for bias correction. • The algorithm effectively overcomes Q-value overestimation and balances exploration-exploitation, enhancing learning stability and control precision in dynamic environments. • Superior control performance is validated across multiple test scenarios, significantly reducing frequency deviation and ACE compared to baseline algorithms. • The method demonstrates strong robustness under varying renewable penetration levels, proving effective for modern power systems with high renewables.

Automatic generation control based on probably approximately correct Bayesian soft actor-critic algorithm with standard Q target

Abstract

Journal

Organization