Earth Science Frontiers ›› 2024, Vol. 31 ›› Issue (3): 530-540.DOI: 10.13745/j.esf.sf.2023.9.56

Previous Articles    

Contribution ratio and distribution patterns of multiple oil sources in the Yanchang Formation of the Ordos Basin: A study utilizing machine learning and interpretability techniques

SU Kaiming1,2(), XU Yaohui1,2,*(), XU Wanglin3, ZHANG Yueqiao3, BAI Bin3, LI Yang1,2, YAN Gang1,2   

  1. 1. Hubei Key Laboratory of Petroleum Geochemistry and Environment, Wuhan 430100, China
    2. College of Resources and Environment, Yangtze University, Wuhan 430100, China
    3. Research Institute of Petroleum Exploration and Development, PetroChina, Beijing 100083, China
  • Received:2023-05-31 Revised:2023-09-06 Online:2024-05-25 Published:2024-05-25

Abstract:

The Yanchang Formation within the Ordos Basin hosts multiple sets of potential source rocks, all exhibiting similar biomarker properties. The conventional method of oil-source correlation has proven ineffective, leading to longstanding debates within the field. In response to these challenges, this study introduces a novel deep learning-based scheme for oil-source comparison, leveraging artificial intelligence methods for research in this domain. The study presents the following key findings and insights: (1) Development of a deep neural network model for identifying the oil source type of unknown samples by utilizing 42 biomarker parameters from a diverse set of mudstone and shale samples representing different oil groups within the Yanchang Formation as training data. The model achieved identification accuracies of 83.0% for Chang 7 mudstone and 79.6% for Chang 8-Chang 10 mudstone, successfully distinguishing the primary source rocks of the Yanchang Formation from hydrocarbon generation products. (2) Application of the model to analyze the oil source classification of numerous sandstone and oil samples. The study calculated the contribution ratios of various source rocks to each oil group within the Yanchang Formation, summarizing their distribution patterns. (3) Conducting sensitivity analysis of the model using the permutation feature importance (PFI) algorithm, revealing differences in biomarkers between the two main source rocks of the Yanchang Formation. These findings contribute to advancing artificial intelligence techniques and technologies in the field of petroleum molecular geochemistry, offering valuable insights for future research and applications.

Key words: machine learning, deep neural network, sensitivity analysis, Yanchang Formation of Ordos Basin, oil-source correlation

CLC Number: