Article Details

Abstract

Accurate identification of fluid properties in low-permeability sandstone reservoirs presents a critical challenge in enhancing exploration efficiency. Conventional well-log interpretation methods often exhibit limited accuracy under conditions such as low resistivity and complex lithology. To address this issue, this study proposes a fluid identification model integrating the MAHAKIL oversampling technique with the XGBoost algorithm. The approach first employs MAHAKIL to mitigate class imbalance in the training dataset and then constructs an XGBoost-based model capable of capturing complex feature relationships for fluid identification. Experimental results demonstrate that data balanced by MAHAKIL significantly improve model performance, with key evaluation metrics such as F1-score and AUC surpassing those of the conventional XGBoost model. These results validate the effectiveness and superiority of the proposed method in handling imbalanced datasets, offering a reliable solution for fluid identification in low-permeability sandstone reservoirs.

Keywords

MAHAKIL oversampling; XGBoost algorithm; low resistivity; fluid identification.

References

1. [1] Ma Xiangdong, Feng Juan, Yin Lu Research on fluid logging identification and reservoir level classification method for low-permeability sandstone reservoirs in the Fushan section of HZ oilfield [J]. Oil and Gas Reservoir Evaluation and Development, 2014, 4 (4): 11-16
2. [2] Hu Wenliang Quantitative identification of fluid properties in low-permeability reservoirs using gas oil ratio calculation based on logging data [J]. Ocean Petroleum, 2023, 43 (2): 48-51
3. [3] Bai Yang, Tan Maojin, Xiao Chengwen, etc Dynamic Classification Committee for Tight Sandstone Gas Reservoirs: Machine Logging Fluid Identification Method [J]. Journal of Geophysics, 2021, 64 (5): 1745-1758
4. [4] Du Ruishan, Huang Yupeng, Fu Xiaofei, etc Identification of Natural Gas Hydrates and Natural Gas Reservoirs Based on SMOTE and XGBoost [J]. Special Oil and Gas Reservoirs, 2024, 31 (5): 11-19
5. [5] Han H，Wang W Y，Mao B H. Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning [C]//International Conference on Intelligent Computing，2005：878-887.
6. [6] Kwabena E B，Jacky K，Passakorn P，et al. MAHAKIL: Diversity Based Oversampling Approach to Alleviate the Class Imbalance Issue in Software Defect Prediction [J].IEEE Transactions on Software Engineering，2018，44（6）：534-550.
7. [7] Ga L A O，Pedro P，Joseph J，et al. The Mahalanobis Distance for Functional Data With Applications to Classification [J].Technometrics，2015，57（2）：281-291.
8. [8] Zhao Weitao, Yin Fuping, Hu Dongchao Preliminary exploration of second-order analytic hierarchy process based on Taylor expansion [J]. Journal of Computational Mechanics, 2022, 39 (5): 655-660
9. [9] Zhang Menghan, Wei Jin, Bian Haiding Machine learning based slope stability analysis method - taking 618 domestic slopes as an example [J]. Journal of Earth Science and Environment, 2022, 44 (6): 1083-1095
10. [10] Yang Jiuqiang, Lin Niantian, Zhang Kai, etc Research on Hyperparameter Selection and Evaluation of Deep Neural Network Models: Taking Multi wave Seismic Response Feature Extraction of Oil and Gas Properties as an Example [J]. Petroleum Geophysics, 2022, 61 (2): 236-244320

Funding

National Major Special Project for New Oil and Gas Exploration and Development "Key Technologies for Ground System Intelligent Operation Paradigm" (2025ZD1408800); Sinopec Zhongyuan Oilfield Branch Technology Project "Data Driven Evaluation of Gas Storage Injection and Production Capacity" (2024BY05)

About the authors

Zhu Gengxue (1989.11-), female, Han nationality, from Jining, Shandong, engineer, master's degree, research direction: exploration and development informatization.