A Model Ensemble Framework Containing a Novel Trial Solution Structure for the Transient Stability Analysis of VSGs Considering Voltage Dynamics

The neural Lyapunov method is a promising tool for the transient stability analysis (TSA) of power systems. However, existing variations suffer from falsification difficulties and inconsistent domain of attraction (DOA) estimations. In this study, we propose i) for the former, a neural Lyapunov function with a novel trial solution structure to reduce the complexity of the falsified function; and ii) for the latter, a TSA framework based on a multi-model ensemble, capable of incorporating various types of candidate Lyapunov functions. The multi-model ensemble framework provides an effective solution to reduce the conservativeness of a single neural Lyapunov function for system DOA estimation, which has the potential to expand the isoenergetic surface defined by the LaSalle lemma. The proposed method was established and evaluated on a virtual synchronous generator model considering voltage dynamics, where the internal voltage is treated as a state variable rather than a parameter. Case studies suggested that the proposed method can effectively deal with strongly coupled dynamical systems with significant time-scale differences in the changes of the state variables. Moreover, compared with other Lyapunov methods, it can less conservatively characterize the stable region.