Measurement-Driven Damping Control Based on the Deep Transfer Reinforcement Learning to Suppress Sub-synchronous Oscillations in a Large-Scale Renewable Power System
Yufan He, Wenjuan Du, Qiang Fu, Haifeng Wang
Abstract
Maintaining power system stability necessitates optimizing power system dynamics and suppressing oscillations. To achieve this objective, significant progress has been made in proposing optimization strategies based on model-based analysis theory. However, accurately obtaining parametric models and operation conditions in a large-scale renewable power system to establish precise analysis models remains challenging. In this paper, we introduce a novel strategy termed Disentangled Factor Transfer Reinforcement Learning (DFTRL) for designing supplementary damping controllers (SDCs) in static synchronous compensators (STATCOM) online to enhance the stability of practical power system. The proposed DFTRL approach allows the reinforcement learning (RL) agent, trained on a simplified power system, to be directly applied to an unseen practical power system. Through case studies, the proposed optimization strategy demonstrates the RL agent's capability to generalize effectively to the target practical power system and successfully suppress oscillations. Moreover, the agent exhibits robustness to variations in power system operating scenarios and noise present in observations.