Hello. I am Zehua Song 👋. My research focuses on using artificial intelligence to solve complex physical system problems.
I received my Master's degree in Artificial Intelligence from China University of Petroleum (Beijing). I was advised by Professor Jin Yang and Associate Professor Yu Song.
My research develops predictive modeling and decision optimization methods for actual engineering problems.
During my graduate studies, I published over 10 papers in international journals and conferences. Six of these were as first author or primary contributor. I also collaborated with companies including CNOOC, CNPC, and Kingsoft Cloud on research projects.
I am the recipient of the Doctoral Research Excellence Assistantship from Texas A&M University, the highest honor for incoming doctoral students, awarded to only 4 recipients across the Texas A&M University System per year. I was also named one of the 100 students nationwide in the People's Daily's list of exemplary National Scholarship recipients.
If you are interested in my work, please feel free to reach out.
My Research Directions
HBPINN: Physics-Informed Bayesian Modeling
HBPINN: Physics-Informed Bayesian Modeling
The HBPINN framework integrates partial differential equations with Bayesian inference. It builds predictive models for complex physical systems with uncertainty quantification capabilities.
DANN + Online Learning: Simulation-to-Reality Model Adaptation
DANN + Online Learning: Simulation-to-Reality Model Adaptation
This is a two-stage model adaptation framework. DANN handles initial domain transfer. Online learning enables continuous improvement. The result is a dynamically updatable model deployable in the field.
Honors and Awards
Check out my latest work
Latent Space Controllable Generation and Inverse Design Tool for Porous Structures
Latent Space Controllable Generation and Inverse Design Tool for Porous Structures
This tool trains a latent space model on voxelized regular tetrahedron porous structure samples to disentangle hole count and pore size features. Given a target hole count and average radius, it searches for candidate structures via latent space sampling and connectivity analysis, enabling inverse generation from macro to micro. It evaluates generation controllability and diversity using structural connectivity and feature matching as objectives.
Drill Bit Dynamics Parameter Inversion and Cumulative Work Analysis Tool
Drill Bit Dynamics Parameter Inversion and Cumulative Work Analysis Tool
This tool uses mud logging data and high-frequency vibration measurements while drilling. It inverts downhole dynamics parameters to quantify energy dissipation from stick-slip, bit bounce, and lateral whirl. It establishes an energy-wear mapping relationship to assess bit cutting structure degradation and remaining useful life. It also recommends optimal drilling parameters for each section to minimize wear and maximize drilling efficiency.
PDC Bit Cutting Structure Parametric Modeling and Optimization Tool
PDC Bit Cutting Structure Parametric Modeling and Optimization Tool
This tool builds a 3D bit model based on blade geometry equations. It uses explicit dynamics simulations to evaluate cutter-rock interactions and local wear characteristics. An implicit wear model extrapolates long-term structural degradation trends. The tool optimizes blade profiles, cutter layouts, and back-rake angles to improve wear distribution and cutting efficiency.
Publications
Under Review
PeSNO: A Physics-embedded Stream Neural Operator as a Warm-Start Preconditioner for Mantle Convection Solvers
PeSNO: A Physics-embedded Stream Neural Operator as a Warm-Start Preconditioner for Mantle Convection Solvers
This paper proposes a Stream Neural Operator for mantle convection simulation. It predicts stream functions to construct divergence-free velocity fields. The method serves as a warm-start preconditioner for numerical solvers. It reduces linear solver iterations by more than half.
Authors: Zehua Song, Yu Song, Jin Yang
Hierarchical Bayesian Physics-Informed Network Model for Open-Hole Wellbore Instability: From Parameter Uncertainty to Reliability Assessment
Hierarchical Bayesian Physics-Informed Network Model for Open-Hole Wellbore Instability: From Parameter Uncertainty to Reliability Assessment
This paper presents a physics-informed network with hierarchical Bayesian inference for open-hole wellbore instability analysis. The model quantifies parameter uncertainty and provides dynamic failure time profiles.
Authors: Zehua Song, Yu Song, Jin Yang, Kejin Chen, Jianqiao Zhang, Xing Wang
Intelligent Prediction and Optimization of Open-Hole Wellbore Multiphysics Stability: A Synergistic PINN-DRL Approach
Intelligent Prediction and Optimization of Open-Hole Wellbore Multiphysics Stability: A Synergistic PINN-DRL Approach
This paper proposes a synergistic framework combining PINN and deep reinforcement learning for wellbore multiphysics stability. It enables high-speed state prediction and drilling fluid parameter optimization.
Authors: Yu Song, Zehua Song, Jin Yang, Kejin Chen, Kun Jiang, Jizhou Tang
Published
Enhancing Energy Efficiency and Sustainability in Offshore Drilling through Real-Time Multi-Objective Optimization: Considering Lag Effects and Formation Variability
Enhancing Energy Efficiency and Sustainability in Offshore Drilling through Real-Time Multi-Objective Optimization: Considering Lag Effects and Formation Variability
This paper presents a multi-objective optimization framework for offshore drilling. It explicitly accounts for formation variability and measurement lag effects. Field data validation demonstrates improved drilling efficiency.
Authors: Yu Song, Zehua Song, Jin Yang, Kejin Chen, Jizhou Tang
A Multi-Objective Reinforcement Learning Framework for Real-Time Drilling Optimization Based on Symbolic Regression and Drilling Perception
A Multi-Objective Reinforcement Learning Framework for Real-Time Drilling Optimization Based on Symbolic Regression and Drilling Perception
This paper combines symbolic regression with deep reinforcement learning for real-time drilling parameter optimization. The method achieves adaptive optimization across different operating conditions.
Authors: Zehua Song, Yu Song, Jin Yang, Baosheng Liu, Bingzhen Gao, Jizhou Tang
A Real-Time Inversion Framework for Carbon Equivalent Emissions in Oil and Gas Extraction Based on Vision Transformer
A Real-Time Inversion Framework for Carbon Equivalent Emissions in Oil and Gas Extraction Based on Vision Transformer
This paper presents a Vision Transformer-based real-time inversion framework for carbon emissions quantification in oil and gas extraction. The framework enables spatial distribution analysis of carbon equivalent emissions.
Authors: Zehua Song, Xiaoyang Yu, Yu Song, Jin Yang, Dongsheng Xu, Kejin Chen, Fangfei Huang, Bin Chen, Yanwei Song
Multi-Objective Optimization Framework for Deepwater Riser Jetting Installation Parameters Using Deep Reinforcement Learning
Multi-Objective Optimization Framework for Deepwater Riser Jetting Installation Parameters Using Deep Reinforcement Learning
This paper presents a deep reinforcement learning optimization framework for deepwater riser jetting installation. It integrates jetting time prediction models with mechanical mechanism analysis.
Authors: Yu Song, Zehua Song, Jin Yang, Lei Li
Early Warning of Deep-Water Drilling Influx Based on Machine Learning
Early Warning of Deep-Water Drilling Influx Based on Machine Learning
This paper develops a machine learning method for early warning of deep-water drilling influx. The approach incorporates bypass line signal enhancement and lag elimination strategies.
Authors: Qishuai Yin, Zehua Song, Kejin Chen, Xu Zhou, Mayank Tyagi, Li Li
A Parzen-Estimator-Enhanced Bayesian Optimization Framework for Deepwater Conductor Jetting
A Parzen-Estimator-Enhanced Bayesian Optimization Framework for Deepwater Conductor Jetting
This paper proposes a segmented Bayesian optimization framework for deepwater conductor jetting process optimization. It combines global search with local refinement strategies.
Authors: Yu Song, Jin Yang, Zehua Song
Co-authored
Deep Learning Based Early Warning Methodology for Gas Kick of Deepwater Drilling Using Pilot-Scale Rig Data
Deep Learning Based Early Warning Methodology for Gas Kick of Deepwater Drilling Using Pilot-Scale Rig Data
Authors: Qishuai Yin, Qikang Zhu, Zehua Song, et al.
Research on Pore Pressure Prediction Technology of HTHP Wells in South China Sea Based on Machine Learning
Research on Pore Pressure Prediction Technology of HTHP Wells in South China Sea Based on Machine Learning
Authors: Dongsheng Xu, Jin Yang, ..., Zehua Song, et al.
Digital Twin: Hybrid Virtual Simulation and Physical Monitoring Strategy for Subsea Wellhead Fatigue Assessment
Digital Twin: Hybrid Virtual Simulation and Physical Monitoring Strategy for Subsea Wellhead Fatigue Assessment
Authors: Kejin Chen, Jin Yang, ..., Zehua Song, et al.
In Preparation
Adaptive Drilling Decision-Making Framework for Extended-Reach Wells under Multiple Objectives
Adaptive Drilling Decision-Making Framework for Extended-Reach Wells under Multiple Objectives
Authors: Yu Song, Zehua Song, ...