Imputation of safety monitoring data for earth-rock dams under sparse data conditions

Safety monitoring data for reservoir dams must be promptly compiled and analyzed to gain real-time insight into dam conditions and to provide essential data for overall safety evaluations. Traditional methods for imputing missing dam data rely on comprehensive pre-existing data and empirical basis functions, which are ineffective for small and medium-sized earth-rock dams with limited data. By utilizing the Empirical Mode Decomposition (EMD) algorithm to analyze missing and homologous monitoring points, effective information can be extracted from minimal data. To address the inconsistency in decomposed components of varying complexity, Dynamic Time Warping (DTW) is employed for clustering and integration. Finally, a prediction model based on the Gated Recurrent Unit (GRU) is built for each clustered dataset, forming an EMD-GRU imputation algorithm for historical monitoring data under sparse data conditions. A comparison between this algorithm and traditional methods, using real-world monitoring data, shows that the mean squared error decreases to below 0.6, demonstrating improved stability and generalization over traditional models in sparse data environments.