Application of artificial intelligence in modeling non-stationary oil filtration in multilayer porous media and data reliability issues

Abstract

This research investigates the modeling of unsteady-state oil filtration processes in heterogeneous multilayer porous media using a hybrid framework that integrates classical fluid dynamics with modern artificial intelligence techniques. The study focuses on developing an efficient computational approach capable of describing fluid transport in complex geological formations characterized by spatial heterogeneity and temporal variability. The governing flow behavior is described through Darcy-based physical laws coupled with continuity equations, while machine learning methods are employed to capture nonlinear dependencies between reservoir parameters such as permeability, porosity, and pressure distribution. In particular, a Physics-Informed Neural Network (PINN) combined with a feedforward neural architecture is utilized to improve prediction accuracy.
In addition, the study addresses critical challenges related to data integrity, uncertainty propagation, and cybersecurity risks in AI-driven simulations. Since model performance strongly depends on the quality of input datasets, ensuring data validation and protection becomes essential. The obtained results demonstrate that hybrid AI–physics models significantly enhance predictive performance and computational efficiency compared to conventional numerical approaches. However, further improvements are required in model interpretability and robustness for real engineering applications.