Deep reinforcement learning framework for adaptive power control in grid-forming inverters: A multi-objective optimization approach DOI
Mrinal Kanti Rajak, Rajen Pudur

Journal of Renewable and Sustainable Energy, Год журнала: 2025, Номер 17(2)

Опубликована: Март 1, 2025

A novel deep reinforcement learning system is introduced, revolutionizing grid-forming inverter control through an attention-based neural architecture with adaptive policy optimization. The uniquely integrates real-time stability constraints multi-objective learning, addressing the fundamental challenges of power under uncertain conditions. approach employs a comprehensive state-space representation incorporating grid dynamics and historical information, complemented by advanced attention mechanism that enables selective feature prioritization across varying operational combines hierarchical network structure prioritized experience replay mechanism, achieving rapid adaptation stable performance. result validation demonstrates improvements over conventional methods, including 43.75% reduction in harmonic distortion (from 3.2% to 1.8%), 46.7% faster dynamic response (8 vs 15 ms), 50% extension operation range weak conditions (operational down short circuit ratio, SCR=1.5). maintains 96% inference accuracy while executing within 50 μ s, meeting requirements. Additionally, superior decoupling performance, reducing coupling effects 80% compared traditional approaches maintaining diverse Learning-based systems electronics demonstrate strong generalization various operating ensuring stability. Integrating opens up new applications for complex problems require adaptability reliability.

Язык: Английский

Deep reinforcement learning framework for adaptive power control in grid-forming inverters: A multi-objective optimization approach DOI
Mrinal Kanti Rajak, Rajen Pudur

Journal of Renewable and Sustainable Energy, Год журнала: 2025, Номер 17(2)

Опубликована: Март 1, 2025

A novel deep reinforcement learning system is introduced, revolutionizing grid-forming inverter control through an attention-based neural architecture with adaptive policy optimization. The uniquely integrates real-time stability constraints multi-objective learning, addressing the fundamental challenges of power under uncertain conditions. approach employs a comprehensive state-space representation incorporating grid dynamics and historical information, complemented by advanced attention mechanism that enables selective feature prioritization across varying operational combines hierarchical network structure prioritized experience replay mechanism, achieving rapid adaptation stable performance. result validation demonstrates improvements over conventional methods, including 43.75% reduction in harmonic distortion (from 3.2% to 1.8%), 46.7% faster dynamic response (8 vs 15 ms), 50% extension operation range weak conditions (operational down short circuit ratio, SCR=1.5). maintains 96% inference accuracy while executing within 50 μ s, meeting requirements. Additionally, superior decoupling performance, reducing coupling effects 80% compared traditional approaches maintaining diverse Learning-based systems electronics demonstrate strong generalization various operating ensuring stability. Integrating opens up new applications for complex problems require adaptability reliability.

Язык: Английский

Процитировано

0