Abstract: To address the challenges of strong energy, significant dispersion, suppression difficulty, and the risk of damaging effective signals caused by guided waves in marine seismic exploration, this study proposes a multi-level guided wave suppression method based on adaptive matching of dispersion characteristics. This method establishes a closed-loop processing workflow consisting of “high-precision dispersion characterization, kinematic forward modeling and prediction, and adaptive matching and subtraction”. First, the nonlinear signal correlation (NLSC) algorithm is introduced to overcome the resolution limitation of conventional dispersion analysis, enabling the detailed characterization of complex multi-mode guided wave energy spectra. Second, a kinematic forward model is constructed based on the extracted dispersion curves to accurately reconstruct the dispersive wavefield of guided waves. Finally, a multi-constrained nonlinear adaptive matched filter is designed to achieve precise subtraction of interference wavefield while preserving effective signals. This method integrates physical constraints with data-driven mechanisms without relying on prior geological models, enabling adaptive modeling and separation of guided waves. Application to real marine seismic data demonstrates that this method effectively resolves the separation difficulty caused by velocity aliasing between guided waves and effective waves in complex shallow-water waveguide environments, significantly improves the signal-to-noise ratio and imaging quality of seismic data, and exhibits strong robustness and high fidelity. It provides an efficient and reliable solution for guided wave suppression in shallow-water seismic exploration.