The integration of grid-connected inverters in large numbers poses challenges to the stability of the power grid. Impedance-based analysis methods are effective means to study the small-signal stability of power electronic systems. However, traditional impedance modeling approaches have limitations when dealing with black-box or grey-box models, especially for grid- connected inverters with unknown control structures and parameters. In this paper, a novel approach is proposed to address the challenges posed by black/grey-box models in grid-connected inverters. The approach is based on a neural network that incorporates physical information and utilizes measured control data. Compared to conventional neural network impedance identification methods, this approach exhibits significantly improved accuracy.