基于模态分解和分类去噪的大坝变形数据清洗方法OA

To address the problem of data noise interference in dam deformation monitoring and improve the quality of monitoring data,a dam deformation data cleaning method based on unsupervised learning is proposed by integrating the grey wolf optimizer(GWO)algorithm,variational mode decomposition(VMD),wavelet threshold denoising(WTD),and the Savitzky-Golay(S-G)filter.In this method,the decomposition parameters of VMD are adaptively optimized using the GWO algorithm,and the optimized VMD is employed to decompose the deformation sequence into multiple intrinsic mode functions(IMFs).Then,the IMFs are classified according to their sample entropy,and classification-based denoising is applied to the high-frequency IMFs and the medium-and low-frequency IMFs.Finally,the processed IMFs are reconstructed to obtain the cleaned deformation data.Case validation on an arch dam shows that the proposed method effectively eliminates noise across multiple frequency bands while well preserving the statistical characteristics of the deformation data,and exhibits robustness to different noise levels.The denoised sequence,when used as training samples,can improve the accuracy of the point-interval prediction model.In addition,the method has good applicability to monitoring sequences of different lengths.