Stability Assessment of Renewable Power System Subsynchronous Oscillation Based on Domain Adversarial Graph Convolutional Network

Large-scale renewable energy systems are prone to subsynchronous oscillations, which will significantly jeopardize system stability if the oscillations cannot be detected timely and accurately. The idea of deep learning (DL) is used to establish a mapping function from steady-state operation data to oscillation classification label, which provides an innovative and efficient technical path for rapid assessment of the oscillation stability in large-scale renewable energy systems. To address the issue of acquiring ample and accurately labeled data for training deep learning models in real-world power systems, this paper proposes a domain adversarial graph convolutional network (DAGCN). It facilitates the transfer of relevant knowledge from simulated simplified power systems to the real-world power systems. Experiments conducted on the modified IEEE 9-Bus system conclusively demonstrate the effectiveness and superiority of the proposed DAGCN method.