Research Article | | Peer-Reviewed

Research on the Application of Bayesian Machine Learning in Reservoir Prediction

Received: 28 October 2023     Accepted: 21 November 2023     Published: 29 November 2023
Views:       Downloads:
Abstract

As exploration and development of various types of oil and natural gas reservoirs underground continue to evolve, the more complex geological environment for seismic exploration in oil and gas reservoirs become focus increasingly. When the target reservoir is vertically interbedded and nested with various lithological strata, and exhibits poor lateral continuity, it becomes increasingly difficult for humans to distinguish oil and gas reservoirs from such complex backgrounds. It's also challenging to quantitatively assess and determine the accuracy of discrimination and achieve optimal reservoir identification results. In response to this issue, the Bayesian machine learning algorithm is introduced for automated target discrimination, enabling efficient differentiation between dolomite reservoirs, mudstones, and other lithological intercalations. The core of applying the Bayesian classifier is to establish a distribution model for target parameters, which is usually assumed to be a known distribution type such as Gaussian or Cauchy distribution. However, in petroleum seismic exploration, the distribution of oil and gas reservoir parameters is highly irregular and significantly different from these established distribution types, limiting the application of the Bayesian classification method. Therefore, we propose using a radial basis function neural network to estimate the prior distribution probability density of oil and gas reservoir parameters. This approach does not assume the prior distribution to be a certain predetermined model but instead builds the prior distribution model based on the numerical distribution characteristics of the target parameters themselves, enhancing the practicality of the Bayesian classification method. This method replaces manual reservoir identification processes, achieving high-precision, quantitative, and automated discrimination of reservoirs. Applied to actual seismic exploration data in oil fields for gas layer prediction, the discrimination results match with industrial gas wells, demonstrating the feasibility and effectiveness of the method.

Published in Petroleum Science and Engineering (Volume 7, Issue 2)
DOI 10.11648/j.pse.20230702.12
Page(s) 35-42
Creative Commons

This is an Open Access article, distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution and reproduction in any medium or format, provided the original work is properly cited.

Copyright

Copyright © The Author(s), 2023. Published by Science Publishing Group

Keywords

Machine Learning, Bayes, Classification, Radial Basis Network, Reservoir Prediction

References
[1] Hartagung R T and Rosid M S. (2022). Implementation of the Poisson Impedance Inversion to Improve Hydrocarbon Reservoir Characterisation in the Poseidon Field, Browse Basin, Australia [J]. Jurnal Penelitian Fisika dan Aplikasinya (JPFA), 12 (2): 102–114.
[2] Suleman Mauritz Sihotang, Ida Herawati1. (2021). Simultaneous Seismic Inversion for Reservoir Characterization at Poseidon Field, Browse Basin, Australia [J]. Journal of Materials Science, Geophysics, Instrumentation and Theoretical Physics, 4 (1): 63-73.
[3] P. S. Nwiyor, E. D. Uko, I. Tamunobereton-ari, et al. (2022). Prediction of Elastic Properties using Crossplot Analysis and Seismic Inversion for Enhanced Quantitative Interpretation of Zeta Field, Niger, Delta, Nigeria [J]. IARJSET, 9 (1): 155-163.
[4] Ashraf, Umar, Hucai Zhang, Aqsa Anees, et al. (2020). Controls on Reservoir Heterogeneity of a Shallow-Marine Reservoir in Sawan Gas Field, SE Pakistan: Implications for Reservoir Quality Prediction Using Acoustic Impedance Inversion [J]. Water 12, no. 11: 2972.
[5] Tom M. Mitchell. (1997). Machine learning [M]. NYC: McGraw-Hill.
[6] Tou J. T., Gonzalez, R. C. (1974). Pattern recognition principles [M], Massachusetts: Addison-Wesley Publishing Company.
[7] Duda R., Hart, P., Stork, D. (2001). Pattern classification [M]. New York: John Wiley & Sons INC.
[8] Bian Z Q, Zhang X G. (1999). Pattern recognition (2nd Edition) [M]. Beijing: Tsinghua University Press. (In Chinese)
[9] Yin X F, Hao Z F, Yang X W. (2007). Review of fast kernel density estimation algorithms [J]. Computer Engineering and Applications, 43 (31): 1-4. (In Chinese)
[10] Greengard L, Strain J. (1991). The fast gauss transform [J]. SIAM J Sci Comput, (2): 79-94.
[11] Gray A G, Moore A W. (2000). N-body problems in statistical learning [J]. Advances in Neural Information Processing Systems, 13 (4): 521-527.
[12] Hall P, Wand M P. (1996). On the accuracy of binned kernel density estimators [J]. Journal of Multivariate Analysis, 56 (2): 165- 187.
[13] Weston J, Gammerman A, Stitson M O, et al. (1999). Support vector density estimation [M]. Cambridge: MIT Press.
[14] Rui T, Zhou Y, Ma g Y, et al. (2011). Target detection using kernel density estimation and Gaussian model cascade mechanism [J]. Computer Engineering and Applications, 47 (18): 1-3. (In Chinese)
[15] Dong M, Zhu H, Xing N, et al. (2012). Calculation method of inverse camera response function of single image based on kernel estimation [J]. Computer Engineering and Applications, 48 (10): 171-174. (In Chinese)
[16] Qiao J F, Zhu H, Shi J, et al. (2012). Fast kernel density estimation method for background modeling [J]. Computer Engineering and Applications, 48 (5): 192-193. (In Chinese)
[17] Avseth P., Mukerji, T., Mavko G. (2005). Quantitative seismic interpretation [M]. New York: Cambirdge University Press, 278-294.
[18] Bertrand, C., Tonellot, T., Fournier, E. (2012). Seismic facies analysis applied to p and s impedances from pre-stack inversion [C]. SEG Expanded Abstract, 72’d Annual Meeting, Salt Lake City, Tulsa, Oklahoma, USA: SEG 2012: 217-220.
[19] P. M. Doyen. (2007). Seismic reservoir characterization: an earth modelling perspective [M]. Netherlands: EAGE.
[20] Chen S C, P. M. Grant, C. F. N. Cowan. (1992). Orthogonal least squares algorithm for radial basis function networks. Radar and signal processing [J]. IEE Proceedings F, 139 (349): 302-309.
[21] Xie W R, Yang W, Li X Z, et al. (2018). The origin and influence of the grain beach reservoirs of Cambrian Longwangmiao formation in central Sichuan area, Sichuan basin [J]. Natural Gas Geoscience, 29 (12): 1715-1726. (In Chinese)
[22] Zhang C, Yang C C, Liu Y C. (2017). Controlling factors of fluid distribution in the lower Cambrian Longwangmiao formation, Moxi area, Sichuan basin [J]. Geology and Exploration, 53 (3): 0599-0608. (In Chinese)
[23] Wang Y P, Yang X F, Wang X Z. (2019). Reservoir property and genesis of powder crystal dolomite in the Longwangmiao formation, Moxi area in central Sichuan basin [J]. Geological Science and Technology Information, 38 (01): 197-205. (In Chinese)
Cite This Article
  • APA Style

    Fu, Z., Chen, K., Liao, J., Long, L., Peng, D. (2023). Research on the Application of Bayesian Machine Learning in Reservoir Prediction. Petroleum Science and Engineering, 7(2), 35-42. https://doi.org/10.11648/j.pse.20230702.12

    Copy | Download

    ACS Style

    Fu, Z.; Chen, K.; Liao, J.; Long, L.; Peng, D. Research on the Application of Bayesian Machine Learning in Reservoir Prediction. Pet. Sci. Eng. 2023, 7(2), 35-42. doi: 10.11648/j.pse.20230702.12

    Copy | Download

    AMA Style

    Fu Z, Chen K, Liao J, Long L, Peng D. Research on the Application of Bayesian Machine Learning in Reservoir Prediction. Pet Sci Eng. 2023;7(2):35-42. doi: 10.11648/j.pse.20230702.12

    Copy | Download

  • @article{10.11648/j.pse.20230702.12,
      author = {Zhiguo Fu and Kang Chen and Juan Liao and Long Long and Da Peng},
      title = {Research on the Application of Bayesian Machine Learning in Reservoir Prediction},
      journal = {Petroleum Science and Engineering},
      volume = {7},
      number = {2},
      pages = {35-42},
      doi = {10.11648/j.pse.20230702.12},
      url = {https://doi.org/10.11648/j.pse.20230702.12},
      eprint = {https://article.sciencepublishinggroup.com/pdf/10.11648.j.pse.20230702.12},
      abstract = {As exploration and development of various types of oil and natural gas reservoirs underground continue to evolve, the more complex geological environment for seismic exploration in oil and gas reservoirs become focus increasingly. When the target reservoir is vertically interbedded and nested with various lithological strata, and exhibits poor lateral continuity, it becomes increasingly difficult for humans to distinguish oil and gas reservoirs from such complex backgrounds. It's also challenging to quantitatively assess and determine the accuracy of discrimination and achieve optimal reservoir identification results. In response to this issue, the Bayesian machine learning algorithm is introduced for automated target discrimination, enabling efficient differentiation between dolomite reservoirs, mudstones, and other lithological intercalations. The core of applying the Bayesian classifier is to establish a distribution model for target parameters, which is usually assumed to be a known distribution type such as Gaussian or Cauchy distribution. However, in petroleum seismic exploration, the distribution of oil and gas reservoir parameters is highly irregular and significantly different from these established distribution types, limiting the application of the Bayesian classification method. Therefore, we propose using a radial basis function neural network to estimate the prior distribution probability density of oil and gas reservoir parameters. This approach does not assume the prior distribution to be a certain predetermined model but instead builds the prior distribution model based on the numerical distribution characteristics of the target parameters themselves, enhancing the practicality of the Bayesian classification method. This method replaces manual reservoir identification processes, achieving high-precision, quantitative, and automated discrimination of reservoirs. Applied to actual seismic exploration data in oil fields for gas layer prediction, the discrimination results match with industrial gas wells, demonstrating the feasibility and effectiveness of the method.
    },
     year = {2023}
    }
    

    Copy | Download

  • TY  - JOUR
    T1  - Research on the Application of Bayesian Machine Learning in Reservoir Prediction
    AU  - Zhiguo Fu
    AU  - Kang Chen
    AU  - Juan Liao
    AU  - Long Long
    AU  - Da Peng
    Y1  - 2023/11/29
    PY  - 2023
    N1  - https://doi.org/10.11648/j.pse.20230702.12
    DO  - 10.11648/j.pse.20230702.12
    T2  - Petroleum Science and Engineering
    JF  - Petroleum Science and Engineering
    JO  - Petroleum Science and Engineering
    SP  - 35
    EP  - 42
    PB  - Science Publishing Group
    SN  - 2640-4516
    UR  - https://doi.org/10.11648/j.pse.20230702.12
    AB  - As exploration and development of various types of oil and natural gas reservoirs underground continue to evolve, the more complex geological environment for seismic exploration in oil and gas reservoirs become focus increasingly. When the target reservoir is vertically interbedded and nested with various lithological strata, and exhibits poor lateral continuity, it becomes increasingly difficult for humans to distinguish oil and gas reservoirs from such complex backgrounds. It's also challenging to quantitatively assess and determine the accuracy of discrimination and achieve optimal reservoir identification results. In response to this issue, the Bayesian machine learning algorithm is introduced for automated target discrimination, enabling efficient differentiation between dolomite reservoirs, mudstones, and other lithological intercalations. The core of applying the Bayesian classifier is to establish a distribution model for target parameters, which is usually assumed to be a known distribution type such as Gaussian or Cauchy distribution. However, in petroleum seismic exploration, the distribution of oil and gas reservoir parameters is highly irregular and significantly different from these established distribution types, limiting the application of the Bayesian classification method. Therefore, we propose using a radial basis function neural network to estimate the prior distribution probability density of oil and gas reservoir parameters. This approach does not assume the prior distribution to be a certain predetermined model but instead builds the prior distribution model based on the numerical distribution characteristics of the target parameters themselves, enhancing the practicality of the Bayesian classification method. This method replaces manual reservoir identification processes, achieving high-precision, quantitative, and automated discrimination of reservoirs. Applied to actual seismic exploration data in oil fields for gas layer prediction, the discrimination results match with industrial gas wells, demonstrating the feasibility and effectiveness of the method.
    
    VL  - 7
    IS  - 2
    ER  - 

    Copy | Download

Author Information
  • School of Geoscience and Technology, Southwest Petroleum University, Chengdu, China

  • Exploration and Development Research Institute, PetroChina Southwest Oil and Gas Field Company, Chengdu, China

  • Southwest Institute of Exploration of Geophysics, BGP Inc., CNPC, Chengdu, China

  • Exploration and Development Research Institute, PetroChina Southwest Oil and Gas Field Company, Chengdu, China

  • Exploration and Development Research Institute, PetroChina Southwest Oil and Gas Field Company, Chengdu, China

  • Sections