Physics-Constrained Bayesian-LSTM Adaptive Impedance Modeling and Resonance Source Identification for Grid-Connected PV-Storage Systems

With the increasing integration of photovoltaic (PV) and energy storage systems into power grids, stability issues caused by converter dynamics, grid impedance interactions, and AC/DC coupling have become increasingly significant. Conventional uniform frequency-scanning methods often fail to provide sufficient resolution in stability-critical frequency regions, limiting the accurate characterization of resonance phenomena. This paper proposes an adaptive two-port admittance modeling and stability assessment method for grid-connected PV-storage systems considering AC/DC coupling characteristics. A two-port frequency-domain model incorporating AC-side admittance, DC-side admittance, and transfer admittance is first established to describe dynamic interactions between the AC and DC subsystems. An adaptive frequency-domain modeling framework with physical constraints is then developed to improve modeling accuracy in critical frequency bands. Furthermore, an improved generalized impedance-ratio criterion considering AC/DC transfer effects and DC-side participation factors is proposed for resonance risk assessment and source identification. Simulation studies and hardware-in-the-loop experiments based on MATLAB 2024b and RT-LAB 2024.1.1 are conducted to validate the proposed method. Results demonstrate that the proposed approach effectively enhances frequency-domain resolution, improves the characterization of AC/DC-coupled resonance behavior, and accurately identifies dominant resonance sources under different operating conditions.