We present a novel reduced order model (ROM) approach for parameterized time-dependent PDEs based on modern learning. The ROM is suitable multi-query problems and nonintrusive. It divided into two distinct stages: nonlinear dimensionality reduction stage that handles the spatially distributed degrees of freedom convolutional autoencoders, time-stepping memory aware neural networks (NNs), specifically causal long short-term NNs. Strategies to ensure generalization stability are discussed. To...

This study investigates the integration of machine learning (ML) and data assimilation (DA) techniques, focusing on implementing surrogate models for Geological Carbon Storage (GCS) projects while maintaining high fidelity physical results in posterior states. Initially, we evaluate modeling capability two distinct models, Fourier Neural Operators (FNOs) Transformer UNet (T-UNet), context CO2 injection simulations within channelized reservoirs. We introduce Surrogate-based hybrid ESMDA...

Localizing leakages in large water distribution systems is an important and ever-present problem. Due to the complexity originating from pipeline networks, too few sensors, noisy measurements, this a highly challenging problem solve. In work, we present methodology based on generative deep learning Bayesian inference for leak localization with uncertainty quantification. A model, utilizing neural serves as probabilistic surrogate model that replaces full equations, while at same time also...

This study investigates the integration of machine learning (ML) and data assimilation (DA) techniques, focusing on implementing surrogate models for Geological Carbon Storage (GCS) projects while maintaining high fidelity physical results in posterior states. Initially, we evaluate modeling capability two distinct models, Fourier Neural Operators (FNOs) Transformer UNet (T-UNet), context CO2 injection simulations within channelized reservoirs. We introduce Surrogate-based hybrid ESMDA...

In the context of solving inverse problems for physics applications within a Bayesian framework, we present new approach, Markov Chain Generative Adversarial Neural Network (MCGAN), to alleviate computational costs associated with inference problem. GANs pose very suitable framework aid in solution problems, as they are designed generate samples from complicated high-dimensional distributions. By training GAN sample low-dimensional latent space and then embedding it Monte Carlo method, can...

This paper presents the development of a Digital Twin (DTwin) to detect and localize leaks in water distribution networks (WDNs), using single-stage two-stage data-driven models. In model, we test anomalies dataset Logistic Regression Random Forest. linear regression model predicts pressure differences between sensor pairs first stage. Based on this, compute residuals. second stage, changes residual are classified Multinomial Forest models possible leak locations' posterior probabilities. We...

Nonlinear model predictive control (NMPC) often requires real-time solution to optimization problems. However, in cases where the mathematical is of high dimension space, e.g. for partial differential equations (PDEs), black-box optimizers are rarely sufficient get required online computational speed. In such one must resort customized solvers. This paper present a new solver nonlinear time-dependent PDE-constrained It composed sequential quadratic programming (SQP) scheme solve problem an...

In Data Assimilation, observations are fused with simulations to obtain an accurate estimate of the state and parameters for a given physical system. Combining data model, however, while accurately estimating uncertainty, is computationally expensive infeasible run in real-time complex systems. Here, we present novel particle filter methodology, Deep Latent Space Particle or D-LSPF, that uses neural network-based surrogate models overcome this computational challenge. The D-LSPF enables...

<title>Abstract</title> In Data Assimilation, observations are fused with simulations to obtain an accurate estimate of the state and parameters for a given physical system. Combining data model, however, while accurately estimating uncertainty, is computationally expensive infeasible run in real-time complex systems. Here, we present novel particle filter methodology, Deep Latent Space Particle or D-LSPF, that uses neural network-based surrogate models overcome this computational challenge....

In data assimilation, observations are fused with simulations to obtain an accurate estimate of the state and parameters for a given physical system. Combining model, however, while accurately estimating uncertainty, is computationally expensive infeasible run in real-time complex systems. Here, we present novel particle filter methodology, Deep Latent Space Particle or D-LSPF, that uses neural network-based surrogate models overcome this computational challenge. The D-LSPF enables filtering...

Download This Paper Open PDF in Browser Add to My Library Share: Permalink Using these links will ensure access this page indefinitely Copy URL DOI

We present a novel reduced order model (ROM) approach for parameterized time-dependent PDEs based on modern learning. The ROM is suitable multi-query problems and nonintrusive. It divided into two distinct stages: A nonlinear dimensionality reduction stage that handles the spatially distributed degrees of freedom convolutional autoencoders, time-stepping memory aware neural networks (NNs), specifically causal long short-term NNs. Strategies to ensure generalization stability are discussed....

In the context of solving inverse problems for physics applications within a Bayesian framework, we present new approach, Markov Chain Generative Adversarial Neural Networks (MCGANs), to alleviate computational costs associated with inference problem. GANs pose very suitable framework aid in solution problems, as they are designed generate samples from complicated high-dimensional distributions. By training GAN sample low-dimensional latent space and then embedding it Monte Carlo method, can...