基于TimeMixer时序预测的电力系统宽频振荡快速辨识OA

Conventional wideband oscillation identification methods for power systems rely exclusively on measured data,resulting in inherent delays that make them unsuitable for the low-latency,real-time monitoring and control requirements of new power systems.To address this limitation,this paper proposes a fast wideband oscillation identification method integrated with TimeMixer time-series prediction.First,a multi-scale decomposition module is employed to generate multi-scale time-series representations suitable for different frequency bands.The past-decomposable mixing(PDM)module then conducts seasonal-trend decomposition and bidirectional cross-scale feature mixing to obtain comprehensive temporal representations.Finally,the future multi-predictor mixing(FMM)module executes independent per-scale prediction and adaptive weighted fusion to produce high-precision short-term future waveforms.By dynamically extending measured data with predicted waveforms,a hybrid measured-predicted time series is constructed,enabling parameter identification to be initiated during the data acquisition process.Consequently,the waiting time for data collection overlaps with the computation time for parameter estimation,significantly reducing the overall identification latency.Simulation results on a power system test network demonstrate that the proposed method substantially reduces the end-to-end latency without sacrificing identification accuracy,thereby significantly improving the real-time capability of online wideband oscillation analysis.