A data enhancement-based quadratic imputation framework for consecutive missing values considering spatiotemporal characteristics of dam deformation

High-quality prototype observations are the basis for a comprehensive analysis of dam structural behavior. However, missing values, especially consecutive missing values (CMVs), are a major barrier. This paper innovatively proposes the concepts of data enhancement (DE) and quadratic imputation (QI) to address CMVs of dam deformation. The Temporal-Spatio Extreme Learning Machine (TsELM) is the core of DE, which exploits the short-term superiority of temporal model to estimate part of the missing segment, providing more reliable modeling information for the imputation of the remaining data. Afterward, the DE components and historical data are incorporated into the training sample and ELM-based QI is performed to obtain the complete simulation results. Analysis shows that the hierarchical imputation method requires fewer parameters and is conducive to constructing a unified imputation framework. Meanwhile, the imputation accuracy of the method is higher than that of traditional models, and it is applicable to dam projects with different deformation sampling frequencies.