Chi Zhang

1. Developed a Physics-Informed Neural Network (PINN) combining CNNs and RNNs using PyTorch to quantify the real-time temporal and spatial evolution of pressure and stress fields in complex reservoir systems. Achieved 97% prediction accuracy and 5x runtime speedup compared to traditional simulators. 2. Built and deployed Generative Adversarial Networks (GANs) to simulate complex stress field dynamics, optimizing performance through multi-physics inputs (e.g., rock formation, fracture configurations). This reduced computational time by 80% and accelerated decision-making in real-time production. 3. Conducted molecular dynamics modeling on multiphase fluid and caprocks, developing quantitative approaches for predicting predicting geochemical and geochemical effects of CO2 and bonded water interactions in CO2 storage process 4. Presented research findings at international conferences, collaborating with sponsors on applying cutting-edge data analysis techniques in the energy industry.

1. Enhanced the resolution of fast-track 4D seismic data processing by developing and comparing various neural network architectures. Implemented evaluation metrics such as SSIM, MS-SSIM, and PSNR to assess the quality of generated images. 2. Designed and implemented deep learning architectures (e.g., cGANs, U-Net, Transformers) using TensorFlow on AWS cloud to enhance 4D seismic imaging, achieving 98% prediction accuracy. 3. Automated data pipelines using Python to integrate seismic datasets with enterprise databases, reducing latency by 90% and saving millions through rapid, accurate AI-driven analyses. 4. Developed physics-informed data augmentation methods to generate synthetic datasets for training deep learning models, improving model robustness and generalization. 5. Improved the prediction accuracy of the designed network by 15% through the integration of multimodal data inputs (e.g., video, image, and time-series data).

1. Developed a physics-informed neural network (based on graph neural network) using PyTorch to optimize well location and control parameters (e.g., injection rate, bottomhole pressure). 2. Predicted temporal evolution of gas saturation and pore pressure with 97% accuracy across 300 generated synthetic waterflooding patterns, achieving a 3x runtime speedup compared to conventional simulators. 3. Incorporated distance-based features into the GNN architecture to quantify producer-injector impacts and handle dynamic variable changes, improving model robustness and interpretability.

1. Developed a deep-learning-based proxy model using TensorFlow for multi-well modeling in discrete fracture reservoirs, achieving 95% prediction accuracy and a 50x runtime speedup on CUDA systems. 2. Collaborated with cross-functional teams to integrate AI solutions into production workflows and designed a user-friendly interface for business units to utilize the proxy for decision-making. 3. Presented findings to senior leadership, demonstrating the model’s impact on operational efficiency and cost savings.

Performed fluid phase behavior and special core flooding experiments including lab QA/QC, routine PVT analyses, and gas EOR lab tests such as swelling and MMP. Lumped pseudo-components and tuned thermodynamic properties to build PVT models in CMG commercial software. Maintained inventory of field sampling equipment and supplies, coordinated with managers, reservoir engineers, and technical advisors regarding projects and sampling jobs.

Simulated the water cut surge after CO2 injection in tight oil reservoirs and matched field data from a pilot test. Several possible mechanisms behind the water cut surge including underestimation of initial water saturation, interfacial tension (IFT) dependent relative permeability, reactivation of water-bearing layers, and re-opening of unpropped hydraulic fractures were investigated