Transient stability quantification of power systems with inverter-based resources via Koopman operator based machine learning approach

Increased integration of inverter-based resources alters the response of large-scale power systems to contingency events. The resulting loss of control actuation and rotating inertia causes the system operating point to move substantially in a short period of time following severe disturbances. To ensure system reliability, it is essential to develop efficient global stability assessment tools. Toward this end, Lyapunov's direct method has received considerable attention due to their rigorous mathematical foundation and fast stability screening. However, most existing approaches in this category are limited in application and cannot readily be extended to practical large-scale power systems. In this work, we propose a data-driven method based on Koopman operator theory for constructing a Lyapunov function and estimating the corresponding region of attraction (ROA). To achieve this, we employ a coordinate transformation enabled by deep neural networks. This approach addresses persistent challenges of existing direct methods in finding proper Lyapunov functions for contemporary power systems. Once the ROA is estimated, the resulting method can rapidly screen the stability of an arbitrary initial operating point without simulating the state trajectory. A numerical case study is presented using a reduced-order model of the North American Western Interconnection with battery energy storage.