Joint identification of hydraulic conductivity and groundwater pollution sources using unscented Kalman smoother with multiple data assimilation and deep learning DOI Creative Commons

Jiuhui Li,

Zhengfang Wu,

Shuo Zhang

и другие.

Ecotoxicology and Environmental Safety, Год журнала: 2025, Номер 295, С. 118134 - 118134

Опубликована: Апрель 1, 2025

Identification of groundwater pollution sources (IGPSs) is a prerequisite for remediation and risk prediction. Data assimilation approaches have been used extensively in IGPSs field recent years. A data approach-unscented Kalman filter complex to operate due the need repeatedly restart simulation model, identification accuracy needs be improved further application with strong nonlinear characteristics. Thus, improve performance enrich technology IGPSs, novel approach called unscented smoother multiple (UKS-MDA) was applied identify hydraulic conductivity GPSs. To assess performance, results (IRs) obtained UKS-MDA were compared those produced by ensemble (ES-MDA) terms computational efficiency. In addition, given learning ability deep belief neural network (DBNN) systems, this study employs as substitute model reduce load loss caused iterative calculations. The indicated that (1) mean relative error (MRE) between DBNN 0.92 %, when it could save approximately 99 % computation time load. (2) MREs IRs using true values scenarios smaller errors concentrations larger 0.4 4.16 lower than ES-MDA. (3) Compared ES-MDA, 12 efficiency execution IGPSs. combination effectively recognize GPSs, which has guiding significance prediction pollution.

Язык: Английский

Joint identification of hydraulic conductivity and groundwater pollution sources using unscented Kalman smoother with multiple data assimilation and deep learning DOI Creative Commons

Jiuhui Li,

Zhengfang Wu,

Shuo Zhang

и другие.

Ecotoxicology and Environmental Safety, Год журнала: 2025, Номер 295, С. 118134 - 118134

Опубликована: Апрель 1, 2025

Identification of groundwater pollution sources (IGPSs) is a prerequisite for remediation and risk prediction. Data assimilation approaches have been used extensively in IGPSs field recent years. A data approach-unscented Kalman filter complex to operate due the need repeatedly restart simulation model, identification accuracy needs be improved further application with strong nonlinear characteristics. Thus, improve performance enrich technology IGPSs, novel approach called unscented smoother multiple (UKS-MDA) was applied identify hydraulic conductivity GPSs. To assess performance, results (IRs) obtained UKS-MDA were compared those produced by ensemble (ES-MDA) terms computational efficiency. In addition, given learning ability deep belief neural network (DBNN) systems, this study employs as substitute model reduce load loss caused iterative calculations. The indicated that (1) mean relative error (MRE) between DBNN 0.92 %, when it could save approximately 99 % computation time load. (2) MREs IRs using true values scenarios smaller errors concentrations larger 0.4 4.16 lower than ES-MDA. (3) Compared ES-MDA, 12 efficiency execution IGPSs. combination effectively recognize GPSs, which has guiding significance prediction pollution.

Язык: Английский

Процитировано

0