When broadband oscillations are analyzed in new energy power stations with large-scale power electronic devices using mechanism-driven models

limitations such as high model order and slow computation speed are encountered

making it difficult to meet the real-time requirements of online dynamic analysis in new electricity system. A data-driven online dynamic analysis model for broadband oscillations in PV power stations based on Koopman theory is thus proposed. Firstly

a high-dimensional linearization mapping of the original system is conducted based on the Koopman operator

and low-dimensional identification of the Koopman operator is performed using a data-driven approach. Then

modal decoupling and identification of oscillation parameters in various frequency bands are achieved through spectral decomposition of the Koopman operator. Eigenvalue distribution is used as the criterion for system stability assessment. Model order reduction is realized by filtering out modes with minor impacts on the system. The contributing factors to oscillations in each frequency band of the PV power station are evaluated using participation factors. Finally

dynamic analysis and parameter updating of the model are carried out based on a sliding data window and closed-loop feedback mechanism

enabling three functions: online identification of broadband oscillation parameters

evaluation of participation factors

and prediction of oscillation states. The results of the case study demonstrated that the proposed model achieves an average accuracy of 96.82% with a total computation time of 0.35 s

outperforming other commonly used analysis methods and meeting the requirements for real-time online analysis.