To effectively mitigate noise interference while fully preserving the original features in dam seepage monitoring data, a denoising method based on optimized SVMD (successive variational mode decomposition) combined with IWT (improved wavelet thresholding) has been proposed, following the “decomposition-screening-denoising-reconstruction” framework. Firstly, SVMD was employed to decompose the seepage monitoring data into a series of modal components, with the penalty factor optimized using BKA (black-winged iris algorithm). Secondly, utilizing MPE (multi-scale permutation entropy) and VCR (variance contribution rate), the obtained modal components were categorized into noise-dominated and signal-dominated groups. The validity of this classification was confirmed through MIC (maximal information coefficient) analysis. Finally, noise-dominated components underwent denoising via IWT and were subsequently reconstructed alongside signal-dominated components to yield the final denoised result. Analyses conducted on both simulated signals and real monitoring data demonstrated that the proposed method significantly enhanced denoising performance by effectively retaining essential seepage signal characteristics while substantially eliminating noise. Furthermore, it exhibited excellent feasibility across all evaluation metrics. This approach serves as a valuable reference for processing dam seepage data.