A Deep Reinforcement Learning-Embedded Two-Stage Framework for Performance-Sensitive Load Frequency Control

• Published in: 2024 IEEE Power & Energy Society General Meeting (PESGM) pp. 1 - 5
• Main Authors: Ma, Yuxin, Hu, Zechun, Zhang, Yong, Wang, Qingwei, Wu, Jiping
• Format: Conference Proceeding
• Language: English
• Published: IEEE 21.07.2024
• Subjects: Deep reinforcement learning; Energy resources; energy storage systems; Frequency control; Inference algorithms; load frequency control; Numerical simulation; Power system reliability; renewable energy resources; Resource management; Robustness; Scalability; Supply and demand
• ISSN: 1944-9933
• DOI: 10.1109/PESGM51994.2024.10689131

Load Frequency Control (LFC) is crucial for maintaining the balance between energy supply and demand in power systems. As emerging flexible resources play increasingly important roles in LFC systems, traditional LFC strategies encounter challenges. While flexible LFC resources exhibit a rapid ramping rate and high response accuracy, their LFC capacities can fluctuate with operational states, sometimes failing to deliver the scheduled LFC capacity. This paper introduces a two-stage performance-sensitive LFC framework to better utilize flexible resources with the above characteristics fully considered. The first stage uses Deep Reinforcement Learning to optimize and allocate LFC commands among heterogeneous resources. The second stage further distributes the commands among homogeneous units within aggregators. In both stages, a performance-based reliable LFC capacity inference is incorporated to address the potential unreliability of flexible LFC resources. Numerical simulations on a single-area LFC system validate the effectiveness of the proposed framework, showcasing improved LFC performance and robustness.