An Improved Tandem Neural Network Architecture for Inverse Modeling of Multicomponent Reactive Transport in Porous Media

Abstract

Reactive transport models (RTMs) are vital tools for understanding solute transport and geochemical reaction processes in subsurface systems and terrestrial environments (

Integration of Deep Learning and Information Theory for Designing Monitoring Networks in Heterogeneous Aquifer Systems

“…In this case, at least N KLE = 100 KLE terms are required to preserve >94% of the field variance (i.e.,$${\sum }_{i=1}^{{N}_{KLE}}{\lambda }_{i}/{\sum }_{i=1}^{\infty }{\lambda }_{i}\approx 0.94$$) (D. Zhang & Lu, 2004; J. Zhang et al., 2020). A parameter with more than 100 dimensions may still cause dimensionality problems (i.e., “curse of dimensionality”) in constructing DNN regression models, that is the computational cost increases exponentially as the input dimensionality increases (S. Mo, Zhu, et al., 2019; Chen et al., 2021). Here, the DCGAN is used to train a generator for creating log‐ k random fields using 10‐dimensional latent vectors obey uniform distribution z ∼ U (−1,1) as input.…”

“…$$\overline{H}$$ is the mean value for all H i ( i = 1,…,M). Specifically, a lower RMSE and a R 2 value approaching 1.0 suggest better performance of DNN regressions (Mo et al., 2020; Chen et al., 2021; Ren et al., 2022). The comparison between the results demonstrates that the constructed DNN regressions may efficiently offer approximate statistical moments and can be integrated for groundwater monitoring network design.…”

Application of different watershed units to debris flow susceptibility mapping: A case study of Northeast China

“…Machine learning models are now widely used because these models can analyze the non-linear corrections between past events and the influencing factors and they predict where disasters will occur (He et al, 2012;Xiong et al, 2020). These models include artificial neural networks (Pham et al, 2017;Chen et al, 2021;Chen et al, 2022), support vector machines (Colkesen et al, 2016), random forest (Hong et al, 2016), decision trees (Althuwaynee et al, 2014), classification and regression tree (Youssef et al, 2015), boosted regression trees (Xiong et al, 2020), Bayesian network (Song et al, 2012), adaptive neuro-fuzzy inference (Jaafari et al, 2019), logistic model tree (Tien Bui et al, 2015) and random gradient descent (Hong et al, 2020). Reichenbach et al (2018) This study compared and analysed the applicability of two different watershed units in regional DFSM based on four models (LR, MLP, CART, and BN).…”

An Efficient Approach for Automatic Complex Fractured Networks Parameter Inversion Based on Surrogate Model and Deep Reinforcement Learning

“…J. Chen et al. (2021) successfully combined tandem neural network architecture with adaptive updating strategy to estimate reactive transport model parameters.…”