Open Access

ARTICLE

A Novel Method for Formation Energy Assessment in Fractured-Vuggy Carbonate Reservoirs Based on the Kolmogorov-Arnold Network

Liang Yan^1,#,*, Yuhui Zhou^1,2,#,*, Xinhui Liu³, Penghua Luo¹, Shaoyang Geng¹, Kaiwen Dong⁴
1 School of Petroleum Engineering, Yangtze University, Wuhan, China
2 Western Research Institute, Yangtze University, Karamay, China
3 PetroChina Xinjiang Oilfield Company No.2 Oil Production Plant, Karamay, China
4 COSL Oilfield Production Research Institute, Tianjin, China
* Corresponding Author: Liang Yan. Email: ; Yuhui Zhou. Email:
# These authors contributed equally to this work
(This article belongs to the Special Issue: Progress and Prospects of Hydraulic Fracture Network Morphology Characterization, Flow Simulation and Optimization Technology for Unconventional Oil and Gas Reservoirs)

Energy Engineering https://doi.org/10.32604/ee.2026.078993

Received 12 January 2026; Accepted 23 March 2026; Published online 20 April 2026

Abstract

Fractured-vuggy carbonate reservoirs are characterised by extreme heterogeneity induced by multi-scale pore-cave-fracture systems, which poses substantial challenges to accurate formation energy assessment—a prerequisite for optimising reservoir development strategies. Traditional methods (e.g., material balance, well-test analysis) are constrained by oversimplified pore space characterisation, failing to quantify the nonlinear impacts of multi-scale media on energy transfer and depletion. To address this critical research gap, this work develops an innovative formation energy evaluation framework that couples the Kolmogorov-Arnold Network (KAN) deep learning architecture with the Whale Optimisation Algorithm (WOA) for adaptive hyperparameter tuning. The proposed model was trained and validated using multi-source geological, engineering, and production datasets from 311 wells in the FM oilfield, Tarim Basin. Comparative results demonstrate that the WOA-KAN model achieves a 42.1% reduction in formation pressure prediction error relative to conventional neural networks, with a final coefficient of determination (R²) of 0.955. SHapley Additive exPlanations (SHAP) analysis identifies cumulative oil production, production time, and large-scale cave development index as the primary drivers of formation energy depletion, while injection-production ratio and natural fracture density exert significant positive effects on energy maintenance. This study provides a data-driven, interpretable tool for dynamic energy assessment in highly heterogeneous carbonate reservoirs, supporting the formulation of efficient energy replenishment strategies.

---

Keywords

Fractured-vuggy carbonate reservoir; formation energy evaluation; whale optimisation algorithm; Kolmogorov-Arnold network; interpretable machine learning

---