Graduation Semester and Year

Fall 2025

Language

English

Document Type

Thesis

Degree Name

Master of Science in Earth and Environmental Science

Department

Earth and Environmental Sciences

First Advisor

Dr. Bezhad Ghanbarian

Second Advisor

Dr. Majie Fan

Third Advisor

Dr. Mohammadreza Soltanian

Abstract

In oil fields, rapid yet reliable production forecasts are crucial for informing operational decisions and maximizing oil recovery. Reservoir simulation is the standard tool for this task, but traditional full-physics models are often computationally expensive and slow, limiting their practicality for quick scenario evaluation. Reduced-order approaches, such as the Capacitance-Resistance Model (CRM), provide faster predictions but often oversimplify reservoir dynamics. To bridge this gap, this study applies a Physics-Informed Neural Network Capacitance-Resistance Model (PINN-CRM) for production forecasting and inter-well connectivity analysis in a water and gas flooded offshore Ghana field with five water injectors, two gas injectors, and eight oil producers.

The PINN-CRM integrates the efficiency of CRM with the rigor of physics-informed neural networks by embedding mass balance, Darcy’s law, and pressure–flow coupling directly into the model training. Using historical injection rates, bottom-hole pressures (BHPs), and time as inputs, the model predicts oil production rates and estimates inter-well connectivity. It is benchmarked against two baselines: a calibrated CRMIP model and a Long Short -Term Memory (LSTM) network. Results show that PINN-CRM outperforms both, achieving a test MAE of 0.04 versus 0.09 for CRMIP and 0.08 for LSTM, while maintaining physical consistency and generalization beyond the training data. Additionally, connectivity maps derived from the PINN-CRM reveal dominant injector–producer relationships, which align well with results from independent field interference tests. This demonstrates the model’s capability not only for accurate forecasting but also for providing diagnostic insights into reservoirs.

Keywords

Capacitance resistance model, Physics-informed neural network, Reservoir simulation, Long Short-Term memory, Inter-well connectivity, Jubilee field

Disciplines

Artificial Intelligence and Robotics | Data Science | Geology | Petroleum Engineering

License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.

Download

Available for download on Wednesday, December 15, 2027

Share

COinS