Abstract:
Deformation is one of the key indicators reflecting the safety status of dams. To enhance the reliability of gross error identification and noise reduction in deformation monitoring data, this study proposes an unsupervised preprocessing method combining the multi-population parallel Rao-1 algorithm, variational mode decomposition (VMD), and multiple discriminative indices. First, the method employs VMD to perform non-recursive decomposition of displacement monitoring sequences at individual measurement points. By introducing average envelope entropy as the objective function, the multi-population parallel Rao-1 algorithm is utilized to optimize VMD hyperparameters, thereby improving the decomposition performance. Subsequently, sample entropy and correlation coefficient indices are applied to separate and locate high-frequency modes containing gross errors and noise features. Finally, gross error identification and noise reduction are achieved through the boxplot method and mode superposition, respectively. Validation using both simulated data and actual horizontal deformation monitoring data of a dam demonstrates that the proposed method excels in gross error localization and noise reduction. It provides novel insights and technical support for the preprocessing of dam deformation monitoring data.