UBHR-Net: An uncertainty-aware band-hypothesis routing network for wideband oscillation monitoring

With the increasing penetration of renewable energy sources, modern power systems are becoming more dynamic and less stable, where oscillations evolve from traditional low-frequency modes to wideband behaviors spanning multiple frequency ranges. As a result, wideband oscillation monitoring requires accurate identification of multiple frequency components and reliable estimation of their dynamic parameters under nonstationary and noisy conditions. However, existing methods often rely on direct feature fusion or fixed temporal modeling, limiting their ability to capture complex oscillatory structures and adapt to varying operating conditions. To address these challenges, this paper proposes an uncertainty-aware band-hypothesis routing network (UBHR-Net) for joint multi-band classification and parameter regression. The proposed framework adopts a dual-branch architecture to extract complementary representations from temporal signals and time–frequency patterns. A band-hypothesis-driven interaction mechanism is introduced to explicitly model candidate oscillatory states and enhance structured feature learning. Furthermore, an uncertainty-aware routing module is designed to adaptively aggregate multiple hypotheses based on their reliability, improving prediction stability under representative noisy and nonstationary conditions. Extensive experiments demonstrate that UBHR-Net outperforms state-of-the-art baselines in both classification accuracy and regression precision, achieving lower RMSE and higher R 2 scores. Ablation and qualitative analyses further verify the effectiveness of each component and the effective dynamic tracking capability of the proposed method. Additional robustness tests, computational-complexity analysis, and a PSCAD-based electromagnetic-transient simulation case further support the stability and practical relevance of the proposed framework. These results indicate that UBHR-Net provides an effective and unified solution for wideband oscillation analysis in complex power systems.