Speaker
Description
Secondary recovery techniques such as water and natural gas injection are extensively applied in the onshore fields of the Algerian Sahara to mitigate reservoir pressure depletion and enhance oil displacement in porous media. However, predicting the evolution of multiphase displacement fronts in heterogeneous quartzitic sandstone reservoirs remains challenging due to strong porosity–permeability contrasts, anisotropy, and the coupled effects of viscous, capillary, and gravity forces.
This study proposes a hybrid physics-informed machine learning framework to characterize and forecast injection-front dynamics and sweep efficiency. The workflow integrates petrophysical well logs, core-scale measurements, pore-scale numerical simulations, and reservoir-scale flow models to construct a multi-scale dataset. Random Forest algorithms are used to infer spatial distributions of porosity and permeability, while a convolutional autoencoder compresses three-dimensional reservoir grids into low-dimensional latent representations.
These latent variables are used to train a surrogate model capable of rapidly predicting breakthrough time, injected-fluid distribution, and cumulative oil recovery at a significantly reduced computational cost compared to conventional reservoir simulations. In parallel, a physics-aware Long Short-Term Memory (LSTM) network is trained on production history data to forecast production rates and water or gas cut.
The results demonstrate that combining physics-based modeling with machine learning improves the prediction of multiphase flow behavior and sweep efficiency, providing an efficient tool for the optimization of injection strategies in heterogeneous sandstone reservoirs.
| References | 1. Aziz, K., & Settari, A. (1979). Petroleum Reservoir Simulation. Applied Science Publishers; 2. Raissi, M., Perdikaris, P., & Karniadakis, G. E. (2019). Physics‐informed neural networks for nonlinear PDEs. Journal of Computational Physics, 378, 686–707; 3. Durlofsky, L. J. (2005). Upscaling and gridding of fine-scale geological models for flow simulation. Computational Geosciences, 9, 259–284; 4. He, J., Reynolds, A. C., & Oliver, D. S. (2018). Reduced-order modeling for subsurface flow simulation using deep learning. Computational Geosciences, 22, 1321–1338; 5. Mo, S., Zhu, Y., Zabaras, N., Shi, X., & Wu, J. (2019). Deep learning for uncertainty quantification of multiphase flow in heterogeneous porous media. Water Resources Research, 55, 703–728. |
|---|---|
| Country | Algérie |
| Acceptance of the Terms & Conditions | Click here to agree |
