基于耗散的动力学感知学习方案用于网络化黑箱构网型逆变器的暂态稳定性分析

Dissipation-based stability analysis, leveraging Lyapunov functions, is popular for assessing the transient stability of inverter-based power systems. However, the proprietary nature of vendor-specific inverter control schemes often results in “black-box” inverter models with limited transparency, while there remains a need for a “generalized” transient stability assessment framework, as opposed to the case-specific designs typically employed in current practice. To address these limitations, this letter proposes a dissipation-based dynamics-aware learning scheme for transient stability analysis of power grids with networked black-box grid-forming inverters. Unlike classical black-box modeling approaches that learn point-wise trajectory mappings as output characteristics, the proposed method directly learns state variable derivatives to uncover the underlying dynamics that govern system evolution. Leveraging the universal approximation capability of neural networks (NNs), the unknown system dynamics are first captured by a NN. Subsequently, data sampled from this NN facilitates the training of a separate NN that represents the Lyapunov function. The proposed bi-level approach enables generalized transient stability evaluation for highly nonlinear system without requiring explicit system models, making it applicable to diverse practical scenarios. Case studies are presented to show the effectiveness of the proposed method.