Sequence impedance identification-based online small-signal stability assessment of grid-connected renewable energy stations considering measurement noise

Small-signal stability is essential for the reliable integration of grid-connected renewable energy stations (RESs). Commercial inverter-based resources (IBRs) within RESs are treated as black boxes due to unknown control structures and parameters, motivating data-driven impedance modeling. However, the accuracy of such methods is impacted by measurement noise. To address this issue, this paper proposes a data-driven framework for online small-signal stability assessment of RESs that develops a deep neural network (DNN) architecture to identify sequence impedance of IBRs while accounting for measurement noise. Specifically, noise originates from two sources: random impedance noise and inherent steady-state measurement noise. During training, accuracy is improved through adaptive sample weighting informed by the noise distribution, which mitigates the impact of noisy impedance data. Adversarial training with virtual perturbations further enhances robustness against noise in steady-state data. This study also reveals distinct impedance characteristics of grid-following (GFL) and grid-forming (GFM) IBRs around the rated frequency and then proposes a classification module to determine the RES type. By selecting the appropriate return-ratio matrix of different RES types, this module can ensure the accuracy of small-signal analysis using simplified Generalized Nyquist Criterion (GNC). Case studies are undertaken on a 6-IBR GFL RES (comprising 1155 samples) and an 8-IBR GFM RES (comprising 2541 samples), with online computation times meeting minute-level monitoring requirements. The proposed method achieves over 30% reduction in magnitude error and 20% reduction in phase-angle error relative to traditional methods for both GFL and GFM cases. Unlike traditional approaches that neglect dual-source noise, simulation results further demonstrate accurate stability determination under marginally unstable operating conditions in the presence of random impedance noise and 40 dB inherent input noise.