Application of ensemble surrogates and adaptive sequential sampling to optimal groundwater remediation design at DNAPLs-contaminated sites

Ouyang, Qiubao; Lu, Wenxi; Miao, Tiansheng; Deng, Wenbing; Jiang, Changlong; Luo, Jiannan

doi:10.1016/j.jconhyd.2017.10.007

Cited by 30 publications

(8 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In previous studies the ensemble data driven models [28,29] have been reported as efficient tools for modeling non-linear problems. A weighted average MM using the optimal weight of single MM can enhance the performance of MM [7]. This technique is based on minimizing the error index (E):…”

Section: Proposed Linear Ensemblesmentioning

confidence: 99%

“…Where m is the weight of each MM, outðxÞ and out ens are the actual value and predicted value of input x respectively. R is the covariance matrix which is expressed as [7]:…”

Section: Proposed Linear Ensemblesmentioning

confidence: 99%

“…In SO model for SWI management, two types of meta-modeling approaches have been generally implemented: offline and online. In the offline approach, the numerical model is completely substituted by MM and sampling algorithm is independent of the optimization model whereas, in the online approach, the MM is trained adaptively using the numerical model during optimization process [7].…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Development of a robust ensemble meta-model for prediction of salinity time series under uncertainty (case study: Talar aquifer)

Ranjbar

Cherubini

2020

Heliyon

View full text Add to dashboard Cite

The aim of this study is to develop an accurate and reliable numerical model of the coastal Talar aquifer threatened by seawater intrusion by developing an ensemble meta-model (MM). In comparison with previous methodologies, the developed model has the following superiority: (1) Its performance is enhanced by developing ensemble MMs using four different meta-modelling frameworks, i.e., artificial neural network, support vector regression, radial basis function, genetic programing and evolutionary polynomial regression; (2) The accuracy of different MMs based on 16 integration of four meta-modeling frameworks is compared; and (3) the effect of aquifer heterogeneity on the MM. The performance of the proposed MM was assessed using an illustrative case aquifer subject to seawater intrusion. The obtained results indicate that the ensemble MM that combines all four meta-modeling frameworks outperformed the GP and ANN models, with a correlation coefficient of 0.98. Moreover, the proposed MM using nonlinear-learning ensemble of SVR-EPR achieves a better and robust forecasting performance. Therefore, it can be considered as an accurate and robust simulator to predict salinity levels under different abstraction patterns in variable density flow. The result of uncertainty analyses reveals that robustness value and pumping rate are inversely proportional and scenarios with a robustness measure of about 12% are more reliable.

show abstract

Section: Proposed Linear Ensemblesmentioning

confidence: 99%

“…Where m is the weight of each MM, outðxÞ and out ens are the actual value and predicted value of input x respectively. R is the covariance matrix which is expressed as [7]:…”

Section: Proposed Linear Ensemblesmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Development of a robust ensemble meta-model for prediction of salinity time series under uncertainty (case study: Talar aquifer)

Ranjbar

Cherubini

2020

Heliyon

View full text Add to dashboard Cite

show abstract

“…Applications on flow and transport problems have been recently investigated to predict contaminant transport and source location (Mo et al., 2019; Wang et al., 2020; X. Yu et al., 2020) as well as optimal well location and pumping schedule for pump‐and‐treat operations (Yan & Minsker, 2006; Yin & Tsai, 2020). Other implementations also included models for the multiphase transport of dense nonaqueous phase liquids (Jiang & Na, 2020; Luo et al., 2020; Ouyang et al., 2017). However, to the best of our knowledge, applications of surrogate models to complex reactive transport problems involving both physical flow and transport processes, as well as comprehensive networks of biogeochemical reactions for natural transport and engineered in situ remediation of contaminated groundwater are still lacking.…”

Section: Introductionmentioning

confidence: 99%

Integrating Process‐Based Reactive Transport Modeling and Machine Learning for Electrokinetic Remediation of Contaminated Groundwater

Sprocati

Rolle

2021

Water Resources Research

View full text Add to dashboard Cite

Advanced reactive transport models of fluid flow and solute transport in subsurface porous media are instrumental for the assessment of contaminant environmental fate and for the design of in situ remediation interventions. However, the increasing complexity of process‐based reactive transport simulators often leads to long runtimes, which poses severe restrictions for tasks that require numerous model evaluations. To overcome this limitation, we demonstrate how machine learning surrogate models, trained on the outputs of a limited number of process‐based reactive transport simulations, can predict the evolution of complex subsurface systems. We focus on electrokinetic enhanced bioremediation of chlorinated solvents in low‐permeability porous media, which is an in situ remediation technology entailing a suite of complex and coupled physical, chemical, and biological processes. A process‐based, multicomponent reactive transport model, capable of describing the key mechanisms of electrokinetic flow and transport, is setup in a two‐dimensional domain. The model accounts for electromigration and electroosmosis, the electrostatic interactions between charged species, the chemistry of the pore water solution, the microbially mediated degradation of the organic compounds, and the dynamics of different degraders. We develop a response surface surrogate framework using an artificial neural network as approximation function and we show that the surrogate model has the capability and the flexibility to capture the complex dynamics of electrokinetic remediation in subsurface porous media and allows computationally efficient model exploration, sensitivity analysis, and uncertainty quantification.

show abstract

“…Because in this work we are interested in approximating an inputoutput relationship, we will focus on data-driven methods. In studies related to subsurface flow and transport, commonly used data-driven methods are the polynomial chaos expansion (e.g., Laloy et al 2013;Wu et al 2014;Zhang et al 2017), support vector machines (e.g., Yoon et al 2011;Wu et al 2015;Xu et al 2017), and Gaussian Process Emulators, which have been used in the modeling of groundwater flow (e.g., Cui et al 2018b, a), unsaturated flow (e.g., Zhang et al 2018;Gadd et al 2019;Zheng et al 2019), subsurface transport (e.g., Ouyang et al 2017;Zhang et al 2018;Gadd et al 2019;Zheng et al 2019), saltwater intrusion (e.g., Rajabi and Ketabchi 2017;Kopsiaftis et al 2019), and processes related to CO 2 -sequestration (e.g., Espinet and Shoemaker 2013;Tian et al 2017;Crevillén-García 2018).…”

Section: Introductionmentioning

confidence: 99%

Sampling behavioral model parameters for ensemble-based sensitivity analysis using Gaussian process emulation and active subspaces

Erdal

Xiao

Nowak

et al. 2020

Stoch Environ Res Risk Assess

View full text Add to dashboard Cite

Ensemble-based uncertainty quantification and global sensitivity analysis of environmental models requires generating large ensembles of parameter-sets. This can already be difficult when analyzing moderately complex models based on partial differential equations because many parameter combinations cause an implausible model behavior even though the individual parameters are within plausible ranges. In this work, we apply Gaussian Process Emulators (GPE) as surrogate models in a sampling scheme. In an active-training phase of the surrogate model, we target the behavioral boundary of the parameter space before sampling this behavioral part of the parameter space more evenly by passive sampling. Active learning increases the subsequent sampling efficiency, but its additional costs pay off only for a sufficiently large sample size. We exemplify our idea with a catchment-scale subsurface flow model with uncertain material properties, boundary conditions, and geometric descriptors of the geological structure. We then perform a global-sensitivity analysis of the resulting behavioral dataset using the active-subspace method, which requires approximating the local sensitivities of the target quantity with respect to all parameters at all sampled locations in parameter space. The Gaussian Process Emulator implicitly provides an analytical expression for this gradient, thus improving the accuracy of the active-subspace construction. When applying the GPE-based preselection, 70–90% of the samples were confirmed to be behavioral by running the full model, whereas only 0.5% of the samples were behavioral in standard Monte-Carlo sampling without preselection. The GPE method also provided local sensitivities at minimal additional costs.

show abstract

Application of ensemble surrogates and adaptive sequential sampling to optimal groundwater remediation design at DNAPLs-contaminated sites

Cited by 30 publications

References 27 publications

Development of a robust ensemble meta-model for prediction of salinity time series under uncertainty (case study: Talar aquifer)

Development of a robust ensemble meta-model for prediction of salinity time series under uncertainty (case study: Talar aquifer)

Integrating Process‐Based Reactive Transport Modeling and Machine Learning for Electrokinetic Remediation of Contaminated Groundwater

Sampling behavioral model parameters for ensemble-based sensitivity analysis using Gaussian process emulation and active subspaces

Contact Info

Product

Resources

About