Volume 67

AI-Powered Pressure Reconstruction and Parameter Inversion for Carbonate Reservoirs: A Physics-Guided Deep Learning Approach Ruofan Yan, Jing Wang, Huiqing Liu

Abstract

Dynamic analysis and reserve estimation for deep fractured-vuggy carbonate reservoirs pose significant challenges due to severe reservoir heterogeneity, complex production dynamics, and the extreme scarcity of key static formation pressure data. While conventional physics-based models struggle with such data sparsity, standard Artificial Intelligence methods often lack the physical consistency and interpretability required for high-stakes engineering applications. To bridge this gap, this paper introduces PG-TCRN, an innovative physics-guided AI architecture that integrates a Temporal Convolutional Network with a Long Short-Term Memory network to synergize deep learning with domain knowledge. Guided by a composite physics-based loss function, this model leverages continuous production data to intelligently reconstruct a reliable formation pressure profile from only a few static measurements. Subsequently, with the material balance equation serving as a strong physical constraint, our framework transforms the history matching task into a low-dimensional optimization problem, enabling the accurate inversion of key parameters like dynamic reserves and water influx coefficient. Application to a real-world case from the Z Reservoir validates our approach, achieving a precise match of the production history and confirming the physical reliability of the inverted parameters. Furthermore, the framework’s interpretable results quantify the contribution of different energy sources, revealing the dominant reservoir drive mechanism. This study presents an efficient and interpretable new paradigm for AI-driven dynamic analysis in fractured-vuggy reservoirs, offering a powerful tool for data-scarce scenarios.

Keywords artificial intelligence, physics-guided deep learning, fractured-vuggy reservoirs, dynamic parameter inversion, intelligent energy systems, material balance equation