Research
University of Texas at Austin
Deep learning-based latent space methods for subsurface resource modeling
Convolutional-Recurrent proxy for spatiotemporal CO2 monitoring
Image-to-video forecasting for geologic CO2 sequestration.
Well placement and control optimization.
Data assimilation and uncertainty quantification.
Latent geologic inversion from multiscale, multi-source dynamic data
Reconstruction of 2D/3D fluvial geologic models from production wells and observation wells data.
Closed-loop forward/inverse modeling.
Residual-Multiscale-Spatiotemporal proxy for H2 storage prediction
Predict H2 plume migration using a spatiotemporal vision transformer.
Dimensionality reduction and parameter estimation from fiber optic data
Estimate injection location and multiphase relative flowrates from fiber optic measurements.
Physics-Informed Anisotropic Resistivity Inversion
Fast inversion algorithm for parallel- and perpendicular-to-bedding-plane resistivity logs.
University of Southern California
Deep learning architectures for data assimilation and proxy models in subsurface energy applications.
Worked in different areas, including: deep neural networks for dimension reduction and data conditioning, novel neural network architectures for characterizing subsurface flow connectivity, integration of dynamic flow data under uncertain geologic scenarios using low-dimensional latent space, and development of efficient fit-for-purpose proxy models for prediction and uncertainty quantification in subsurface flow systems.
Deep Learning-based Scale Relationships in Reservoir Simulation: development of a regression proxy model for reducing coarse-scale errors in a 2D waterflooding application; deep learning architectures for coarse-to-fine scale mapping of static and dynamic data.
University of Tulsa
Boundary Element Method for the Dirichlet Problem for Laplace's Equation on a Disk
Created a Mathematica program to evaluate any function satisfying the Dirichlet problem for elliptical partial differential equations on a circular domain. Linear, piece-wise elements were used for the boundary element method solution. Visualization, numerical analysis and error computations were performed on the results.
Predictive Analytics and Artificial Lift Solution for Liquid Loading wells using Statistical Learning and Physical Modeling
Worked in collaboration with Bravo Natural Resources in developing an algorithm to predict liquid loading in gas wells in the STACK/SCOOP region. Used R/Shiny to deploy a real-time user-friendly GUI, "LoadLook," for visualization and prediction of production performance. Used ARIMA and modified decline curve analysis for liquid loading prediction, and cross-validation with mechanistic physical modeling.
