Kolmogorov–Arnold-Network-Based Impedance Identification of Power Electronic Systems With Grid-Following and Grid-Forming Control

Impedance model is being adopted for the stability analysis of power electronic systems. Due to limited accessibility to detailed controller information, the analytical impedance model is difficult to derive. This letter proposes an impedance identification method based on Kolmogorov-Arnold Network (KAN) and physical information for grid-connected converter systems with grid-following (GFL) and grid-forming (GFM) control. Leveraging the Kolmogorov–Arnold representation theorem, the proposed two-layer KAN replaces fixed neuronal weights with learnable univariate spline functions, thereby universal approximation with fewer trainable parameters and reduced network depth than multi-layer perceptron (MLP) with comparable accuracy is achieved. The effectiveness of the proposed method is validated by wind farms with doubly fed induction generator (DFIG) converters in RTDS and compared with other neural structures.