A Power Regulation Strategy for Grid-Forming Wind Power Clusters Based on Deep Reinforcement Learning
- 1School of Mechanical and Electrical Engineering, Henan Institute of Science and Technology, Xinxiang 453003, China
Abstract
Renewable-rich power systems rely on converter-interfaced resources, reducing synchronous inertia and complicating frequency regulation. Grid-forming wind turbines can provide active-power support, but cluster-level commands must respect rotor-speed and DC-link voltage limits. This study develops a deep-reinforcement-learning power regulation strategy for grid-forming wind power clusters using a safety-constraint-integrated dual-layer Soft Actor–Critic algorithm. A coupled dynamic model of the regional grid, synchronous generator, wind-turbine grid-side converter, drivetrain, and DC link is established to characterize the active power–frequency response and unit operating constraints. The upper layer determines the cluster regulation demand, whereas the lower layer allocates commands among turbines and maps preliminary actions to componentwise operating bounds. A MATLAB/Simulink case study containing one 60 MW synchronous generator and twenty 2 MW grid-forming wind turbines compares the proposed strategy with traditional automatic generation control over 50 disturbance realizations. The proposed strategy reduced frequency standard deviation from 17.5 to 14.2 mHz, improved cumulative reward from −4.76×107 to −3.84×107, and increased safety-constraint satisfaction from 82.1% to 99.2%. The results support hierarchical reinforcement learning for coordinating fast frequency support while improving adherence to turbine operating limits.
Related
- Constrained Reinforcement Learning-Based Framework for DC Microgrid: Current Sharing and Voltage Regulation
- Dynamic Multi-Pair Trading Strategy in Cryptocurrency Markets with Deep Reinforcement Learning
- An Improved Escape Algorithm with Reinforcement Learning and Multi-Strategy Synergy for Global Optimization and Real Problems
- Reinforcement Learning Under Runtime Confounding: Policy Learning and Off-Policy Evaluation
- Deep Reinforcement Learning for Blind Passive Beamforming in RIS-Assisted MIMO Systems
BibTeX
@article{zhu2026power,
title = {{A Power Regulation Strategy for Grid-Forming Wind Power Clusters Based on Deep Reinforcement Learning}},
author = {Yuelun Zhu and Xiaohong Kong},
journal = {Mathematics},
year = {2026},
month = sep,
doi = {10.3390/math14193554},
url = {https://www.mdpi.com/2227-7390/14/19/3554},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A Power Regulation Strategy for Grid-Forming Wind Power Clusters Based on Deep Reinforcement Learning
AU - Yuelun Zhu
AU - Xiaohong Kong
PY - 2026
DA - 2026/09/30
JO - Mathematics
DO - 10.3390/math14193554
UR - https://www.mdpi.com/2227-7390/14/19/3554
AB - Renewable-rich power systems rely on converter-interfaced resources, reducing synchronous inertia and complicating frequency regulation. Grid-forming wind turbines can provide active-power support, but cluster-level commands must respect rotor-speed and DC-link voltage limits. This study develops a deep-reinforcement-learning power regulation strategy for grid-forming wind power clusters using a safety-constraint-integrated dual-layer Soft Actor–Critic algorithm. A coupled dynamic model of the regional grid, synchronous generator, wind-turbine grid-side converter, drivetrain, and DC link is established to characterize the active power–frequency response and unit operating constraints. The upper layer determines the cluster regulation demand, whereas the lower layer allocates commands among turbines and maps preliminary actions to componentwise operating bounds. A MATLAB/Simulink case study containing one 60 MW synchronous generator and twenty 2 MW grid-forming wind turbines compares the proposed strategy with traditional automatic generation control over 50 disturbance realizations. The proposed strategy reduced frequency standard deviation from 17.5 to 14.2 mHz, improved cumulative reward from −4.76×107 to −3.84×107, and increased safety-constraint satisfaction from 82.1% to 99.2%. The results support hierarchical reinforcement learning for coordinating fast frequency support while improving adherence to turbine operating limits.
SN - 2227-7390
ER -