Scholarlink

The race for energy becomes more challenging every day. In the case of oil and gas reservoirs, most near-surface fields have already been depleted. Consequently, discovering new reservoirs necessitates the exploration of deep areas characterized by limited knowledge and high uncertainty. The quantification of this uncertainty is crucial for making informed decisions. Simultaneously, the oil and gas industry has experienced a notable increase in utilizing machine learning tools, particularly for facies classification. In this paper, we delve into the applicability of these tools to assess the uncertainty in the initial phases of oil and gas reservoir exploration. This uncertainty is defined by multiple geological scenarios that may elucidate the subsurface geology. The methodology outlined here evaluates the probability of each geological scenario using machine learning algorithms. The process starts with establishing a space of uncertainty, considering all geological scenarios through geostatistical simulations. These simulations serve as input for a transfer function that links them to the available data. In this instance, geostatistical simulations are employed as input for rock physics and forward model functions to generate simulations of seismic amplitudes. These simulated seismic amplitudes and the true seismic amplitude are represented in a multidimensional scaling (MDS). Finally, the probability of each geological scenario is determined based on the distance between the simulations of that scenario and the actual seismic data in the MDS space. To obtain these probabilities, we compared the performance of four machine learning algorithms: (i) K-Nearest Neighbor, (ii) Random Forest, (iii) Support Vector Classifiers, and (iv) Gaussian Naïve Bayes. The results demonstrated a general concordance among the methods, facilitating the identification of the most probable geological scenario and highlighting the applicability of machine learning algorithms in quantifying uncertainty during the early stages of oil and gas reservoir exploration.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Uncertainty Evaluation of Geological Scenarios in Early Oil Exploration Stages by Machine Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • W. Emilio G. Moreno,
  • Marcel Antônio Bassani,
  • João Felipe Coimbra Leite Costa,
  • João Lucas de Oliveira Alves,
  • Leonardo Azevedo,
  • Diego Marques,
  • Amílcar Soares,
  • Rodrigo Peroni,
  • Vanessa Koppe

摘要

The race for energy becomes more challenging every day. In the case of oil and gas reservoirs, most near-surface fields have already been depleted. Consequently, discovering new reservoirs necessitates the exploration of deep areas characterized by limited knowledge and high uncertainty. The quantification of this uncertainty is crucial for making informed decisions. Simultaneously, the oil and gas industry has experienced a notable increase in utilizing machine learning tools, particularly for facies classification. In this paper, we delve into the applicability of these tools to assess the uncertainty in the initial phases of oil and gas reservoir exploration. This uncertainty is defined by multiple geological scenarios that may elucidate the subsurface geology. The methodology outlined here evaluates the probability of each geological scenario using machine learning algorithms. The process starts with establishing a space of uncertainty, considering all geological scenarios through geostatistical simulations. These simulations serve as input for a transfer function that links them to the available data. In this instance, geostatistical simulations are employed as input for rock physics and forward model functions to generate simulations of seismic amplitudes. These simulated seismic amplitudes and the true seismic amplitude are represented in a multidimensional scaling (MDS). Finally, the probability of each geological scenario is determined based on the distance between the simulations of that scenario and the actual seismic data in the MDS space. To obtain these probabilities, we compared the performance of four machine learning algorithms: (i) K-Nearest Neighbor, (ii) Random Forest, (iii) Support Vector Classifiers, and (iv) Gaussian Naïve Bayes. The results demonstrated a general concordance among the methods, facilitating the identification of the most probable geological scenario and highlighting the applicability of machine learning algorithms in quantifying uncertainty during the early stages of oil and gas reservoir exploration.