Regression-tree-based data-efficient parameterized admittance modeling and stability analysis of grid-connected inverters

Accurate knowledge of the terminal admittance of three-phase grid-connected inverters is essential for reliable stability assessment. However, the limited availability of internal control and circuit parameters from manufacturers makes analytical admittance modeling challenging, motivating the use of data-driven approaches. This paper proposes a regression tree–based framework to estimate inverter admittance directly from operating conditions. In contrast to existing machine learning (ML) methods that rely on operating conditions and frequency-sweep inputs and predict admittance in a point-wise frequency-domain form (e.g., magnitude and phase), the proposed approach estimates the coefficients of a parameterized transfer-function model. This formulation inherently captures frequency-dependent behavior, eliminating the need for frequency as an additional input and reducing input dimensionality and computational complexity, thereby improving data efficiency. Moreover, the proposed regression tree provides an interpretable structure that explicitly links operating conditions to admittance characteristics. The predicted admittance is further utilized for stability analysis by deriving admissible regions of grid impedance parameters, including resistance and inductance. Case studies across multiple operating points demonstrate accurate admittance reconstruction and consistent stability assessment. In the low-frequency range, the proposed method achieves a lower average error compared to a feedforward neural network (NN) benchmark, supporting its effectiveness.