This repository presents the implementation of a Grid-Based Smart Proxy model for reservoir simulation using Artificial Neural Networks (ANN).
The objective is to reproduce reservoir pressure and saturation distribution during CO₂ injection processes.
The methodology follows a data-driven reservoir modeling approach.
Reservoir simulation data are used to train Artificial Neural Networks capable of approximating the results of full-physics simulation models.
Workflow:
- Reservoir simulation data generation
- Data preprocessing
- Feature engineering
- ANN training
- Model validation
The dataset is constructed using reservoir simulation outputs and includes spatial and temporal information for each grid cell.
Main features:
- Grid location (i, j, k)
- Injection rate
- Reservoir properties
- Pressure
- Saturation
- Time step
The Smart Proxy model is based on Artificial Neural Networks.
Example architecture:
- Input layer: reservoir features
- Hidden layers: fully connected layers
- Activation: Sigmoid
- Output layer:
- Pressure prediction
- Saturation prediction
The trained Smart Proxy model successfully reproduces the dynamic behavior of the reservoir simulation.
The proxy significantly reduces computational time compared to traditional numerical simulators while maintaining acceptable accuracy.
The data for this project was obtained through this link: SHAHKARAMI A. Database example for SACROC case study. (2019-07)[2019-07-02]. https://www.researchgate.net/publication/334139922_Sample_of_database_for_the_case_study_of_SCAROC_CO2_EOR_and_Storage?channel=doi&linkId=5d1a2e5aa6fdcc2462b69ae2&showFulltext=true
SHAHKARAMI Alireza1,*, MOHAGHEGH Shahab2. Applications of smart proxies for subsurface modeling. PETROL. EXPLOR. DEVELOP., 2020, 47(2): 400–412.