Open Access

ARTICLE

Deep Learning-Based Prediction of Seepage Flow in Soil-Like Porous Media

Zhenzhen Shen^1,2, Kang Yang², Dengfeng Wei², Quansheng Liang², Zhenpeng Ma², Hong Wang², Keyu Wang², Chunwei Zhang², Xiaohu Yang^3,*

1 Department of Geology, Northwest University, 229 Taibai North Road, Xi’an, 710069, China
2 Shaanxi Yanchang Petroleum (Group) Research Institute, No. 75 Keji 2nd Road, High-Tech Zone, Xi’an, 710075, China
3 Institute of the Building Environment & Sustainability Technology, School of Human Settlements and Civil Engineering, Xi’an Jiaotong University, Xi’an, 710049, China

* Corresponding Author: Xiaohu Yang. Email:

Fluid Dynamics & Materials Processing 2025, 21(11), 2741-2760. https://doi.org/10.32604/fdmp.2025.070395

Received 15 July 2025; Accepted 06 November 2025; Issue published 01 December 2025

Abstract

The rapid prediction of seepage mass flow in soil is essential for understanding fluid transport in porous media. This study proposes a new method for fast prediction of soil seepage mass flow by combining mesoscopic modeling with deep learning. Porous media structures were generated using the Quartet Structure Generation Set (QSGS) method, and a mesoscopic-scale seepage calculation model was applied to compute flow rates. These results were then used to train a deep learning model for rapid prediction. The analysis shows that larger average pore diameters lead to higher internal flow velocities and mass flow rates, while pressure drops significantly at the throats of fine pores. The trained model predicts seepage mass flow rates with deviations within ±20%, achieving a root mean square error of 0.24261 and an average deviation of –0.02197. Importantly, the method performs well even with limited training data, though image-based deep learning approaches may yield better accuracy when larger datasets are available.

---

Keywords

---