scispace - formally typeset
Search or ask a question
Journal ArticleDOI

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

10 Nov 2021-Water Resources Research (John Wiley & Sons, Ltd)-Vol. 57, Iss: 12
TL;DR: The results show that TNNA‐AUS successfully reduces the inversion bias and improves the computational efficiency and inversion accuracy, compared with the global improvement strategy of adding training samples according to the prior distribution of model parameters.
About: This article is published in Water Resources Research.The article was published on 2021-11-10 and is currently open access. It has received 25 citations till now. The article focuses on the topics: Deep learning.
Citations
More filters
Journal ArticleDOI
TL;DR: In this paper , the authors developed an integrated inversion framework, which combines the geological single-sample generative adversarial network (GeoSinGAN), the deep octave convolution dense residual network (DOCRN), and the iterative local updating ensemble smoother (ILUES), named GeoSinGAN-DOCRN-ILUES, for more efficiently generating heterogeneous aquifer structures.
Abstract: Generating reasonable heterogeneous aquifer structures is essential for understanding the physicochemical processes controlling groundwater flow and solute transport better. The inversion process of aquifer structure identification is usually time-consuming. This study develops an integrated inversion framework, which combines the geological single-sample generative adversarial network (GeoSinGAN), the deep octave convolution dense residual network (DOCRN), and the iterative local updating ensemble smoother (ILUES), named GeoSinGAN-DOCRN-ILUES, for more efficiently generating heterogeneous aquifer structures. The performance of the integrated framework is illustrated by two synthetic contaminant experiments. We show that GeoSinGAN can generate heterogeneous aquifer structures with geostatistical characteristics similar to those of the training sample, while its training time is at least 10 times faster than that of typical approaches (e.g., multi-sample-based GAN). The octave convolution layer and multi-residual connection enable the DOCRN to map the heterogeneity structures to the state variable fields (e.g., hydraulic head, concentration distributions) while reducing the computational cost. The results show that the integrated inversion framework of GeoSinGAN and DOCRN can effectively and reasonably generate the heterogeneous aquifer structures.

29 citations

Journal ArticleDOI
TL;DR: Wang et al. as discussed by the authors developed a novel upscaling framework to predict roof water inflow by integrating the multiscale hydrogeological properties of roof aquifers, which can guide the development of methods for considering micropores and fractures simultaneously.
Abstract: Underground coal mining suffers from groundwater intrusion from the aquifers overlying coal seams. Therefore, developing methods for the accurate prediction of roof water inflow is urgently needed to design a safe drainage system. In this study, we developed a novel upscaling framework to predict roof water inflow by integrating the multiscale hydrogeological properties of roof aquifers. In this framework, we imaged rock samples via scanning electron microscopy and performed pore-scale analysis based on fractal theory. A fractal model of permeability was introduced to calculate the seepage capacity of the pore structure in the samples. The effect of fractures was further evaluated via core-scale pneumatic experiments. Subsequently, we derived an upscaling formula of hydraulic conductivity used for predicting roof water inflow at the field scale. The proposed upscaling approach was demonstrated using data from a coal mine in Northern China. The results indicate that the actual water inflow (21 m3/h) is within the predicted range of our upscaling framework (9.32–92.78 m3/h), and the initial line fracture rate dx is distributed between 0.02 % and 0.03 %. Therefore, these findings can guide the development of methods for considering micropores and fractures simultaneously and scaling them up to the field scale for effective prediction of water inflow from roof aquifers.

19 citations

Journal ArticleDOI
TL;DR: In this article , an integrated framework is developed to guide the monitoring network optimization and duration selection for solute transport in heterogeneous sand tank experiments, which is designed based on entropy and data worth analysis.
Abstract: This study develops an integrated framework to guide the monitoring network optimization and duration selection for solute transport in heterogeneous sand tank experiments. The method is designed based on entropy and data worth analysis. Numerical models are applied to approach prior observation datasets and to support optimization analysis. Several candidate monitoring locations are synthetically assumed in numerical models. Entropy analysis considers local scale heterogeneity in experiment and identifies stable monitoring locations through extracting maximum information and minimizing optimization redundancies. Data worth analysis quantifies the potential of observation data to reduce the uncertainty of key parameters and selects the monitoring locations with higher data worth. Final monitoring network comprises of optimized monitoring locations obtained based on entropy and data worth analysis. A lab-scale tracer experiment is presented to explore the applicability of the proposed framework. Results show that the optimized monitoring network can accurately characterize the distribution of contaminant plumes in 3D domains and provides estimation of key flow and transport parameters (e.g., hydraulic conductivity and dispersivity). With the extension of experiment time, the total information of monitoring network is maximized, while the uncertainty of key parameters is minimized. The recommended experimental duration is the time by which both joint entropy and parameter variation coefficients are stabilized. Our developed methodology can be used as a flexible and powerful tool to design more complex transport experiments at different spatiotemporal scales.

18 citations

DOI
TL;DR: In this paper , the authors extended a recently developed stage-wise stochastic deep learning inversion framework by coupling it with nonisothermal flow and transport simulations to estimate subsurface structures.
Abstract: Reliable characterization of subsurface structures is essential for earth sciences and related applications. Data assimilation‐based identification frameworks can reasonably estimate subsurface structures using available lithological (e.g., borehole core, well log) and dynamic (e.g., hydraulic head, solute concentration) observations. However, a reasonable selection of the observation type and frequency is essential for accurate structure identification. To achieve this, we extended a recently developed stage‐wise stochastic deep learning inversion framework by coupling it with non‐isothermal flow and transport simulations. With the extended framework, the worth of three common observations (hydraulic head, concentration, and temperature) are compared under different observation noise and frequency. The framework combines the emerging deep‐learning (DL)‐based framework with the traditional stochastic approaches. This combination makes it possible to simultaneously compare the ability of these two methods to assimilate observation data. Our results show that including at least one type of dynamic observation strongly improves subsurface structure identifiability and reduces the uncertainty. However, the DL‐based framework is able to identify subsurface structures more accurately than stochastic identification methods under the same scenarios. Assimilation of certain types of dynamic observations could reduce the prediction error for related dynamic responses, but not necessarily for other uncorrelated dynamic responses. Observation data worth is affected by the observation noise and frequency. High observation noise increases the uncertainty of the prediction and reduces the estimation accuracy. However, the higher observation frequency can significantly improve the temporal dynamic information of observations. This information can compensate for negative impacts of high observation noise.

17 citations

References
More filters
Proceedings ArticleDOI
27 Jun 2016
TL;DR: In this article, the authors proposed a residual learning framework to ease the training of networks that are substantially deeper than those used previously, which won the 1st place on the ILSVRC 2015 classification task.
Abstract: Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers—8× deeper than VGG nets [40] but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers. The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions1, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

123,388 citations

Journal ArticleDOI
28 May 2015-Nature
TL;DR: Deep learning is making major advances in solving problems that have resisted the best attempts of the artificial intelligence community for many years, and will have many more successes in the near future because it requires very little engineering by hand and can easily take advantage of increases in the amount of available computation and data.
Abstract: Deep learning allows computational models that are composed of multiple processing layers to learn representations of data with multiple levels of abstraction. These methods have dramatically improved the state-of-the-art in speech recognition, visual object recognition, object detection and many other domains such as drug discovery and genomics. Deep learning discovers intricate structure in large data sets by using the backpropagation algorithm to indicate how a machine should change its internal parameters that are used to compute the representation in each layer from the representation in the previous layer. Deep convolutional nets have brought about breakthroughs in processing images, video, speech and audio, whereas recurrent nets have shone light on sequential data such as text and speech.

46,982 citations

Proceedings Article
01 Jan 2014
TL;DR: A stochastic variational inference and learning algorithm that scales to large datasets and, under some mild differentiability conditions, even works in the intractable case is introduced.
Abstract: How can we perform efficient inference and learning in directed probabilistic models, in the presence of continuous latent variables with intractable posterior distributions, and large datasets? We introduce a stochastic variational inference and learning algorithm that scales to large datasets and, under some mild differentiability conditions, even works in the intractable case. Our contributions is two-fold. First, we show that a reparameterization of the variational lower bound yields a lower bound estimator that can be straightforwardly optimized using standard stochastic gradient methods. Second, we show that for i.i.d. datasets with continuous latent variables per datapoint, posterior inference can be made especially efficient by fitting an approximate inference model (also called a recognition model) to the intractable posterior using the proposed lower bound estimator. Theoretical advantages are reflected in experimental results.

20,769 citations


"An Improved Tandem Neural Network A..." refers background in this paper

  • ...…low-dimensional latent space to high-dimensional space-dependent parameters through some unsupervised learning methods, including generative adversarial network (Goodfellow et al., 2014), variational autoencoder (Kingma & Welling, 2013), and their variants (Cheng et al., 2020; Mo et al., 2020)....

    [...]

Journal ArticleDOI
TL;DR: In this article, the authors introduce physics-informed neural networks, which are trained to solve supervised learning tasks while respecting any given laws of physics described by general nonlinear partial differential equations.

5,448 citations


Additional excerpts

  • ...He et  al., 2020; Meng & Karniadakis, 2020; Raissi et  al., 2019; L. Sun et  al., 2019; N. Wang, et  al., 2020)....

    [...]

Journal ArticleDOI
TL;DR: This work represents the stochastic processes with an optimum trial basis from the Askey family of orthogonal polynomials that reduces the dimensionality of the system and leads to exponential convergence of the error.
Abstract: We present a new method for solving stochastic differential equations based on Galerkin projections and extensions of Wiener's polynomial chaos Specifically, we represent the stochastic processes with an optimum trial basis from the Askey family of orthogonal polynomials that reduces the dimensionality of the system and leads to exponential convergence of the error Several continuous and discrete processes are treated, and numerical examples show substantial speed-up compared to Monte Carlo simulations for low dimensional stochastic inputs

4,473 citations


"An Improved Tandem Neural Network A..." refers background in this paper

  • ...This includes and is not limited to polynomial chaos expansion (Laloy et al., 2013; Xiu & Karniadakis, 2002), Gaussian process (H. Wang & Li, 2018; J. Zhang et al., 2016), Kriging surrogate modeling (X. Yan et al., 2019; J. Zhou et al., 2018), support vector machine (Lal & Datta, 2018; Xingpo…...

    [...]