Scholarlink

<p>Accurate prediction of the ground stress field is an important scientific issue in the evaluation of the compressibility of coalbed methane reservoirs and the design of hydraulic fracturing. This study addresses the challenges of multi-source data fusion and insufficient generalization capability for small samples in predicting in-situ stress in complex coal reservoirs. It proposes an in-situ stress prediction model ANFIS-PSO that integrates the Particle Swarm Optimization (PSO) algorithm with an Adaptive Neuro-Fuzzy Inference System (ANFIS). The model integrates multi-source data including acoustic time-of-flight logging, measured in-situ stress values from injection/depression tests, and Brazilian splitting and uniaxial compression experiments. Through Pearson correlation analysis (|R|≥ 0.31, <i>p</i> &lt; 0.001) to identify five key input parameters: GR, LLS, AC, DEN, and TVD. This established a nonlinear mapping relationship between logging responses and in-situ stress. The PSO algorithm was employed to globally optimize the Gaussian membership function parameters and fuzzy rule weights of the ANFIS model. After 228 iterations, the model's convergence and accuracy were significantly enhanced. The results show that: compared with the MLP-PSO and RF-PSO, the RMSE of the ANFIS-PSO for predicting the minimum horizontal principal stress (<i>σ</i><sub><i>h</i></sub>) decreases by 25.0% and 39.2% respectively; for predicting the maximum horizontal principal stress (<i>σ</i><sub><i>H</i></sub>), the RMSE decreases by 31.8% and 72.7% respectively, and the coefficient of determination R<sup>2</sup> ranges from 0.96 to 0.97 (n = 216). The SHAP interpretability analysis indicates that the contribution of true vertical depth (TVD) to <i>σ</i><sub><i>h</i></sub> and <i>σ</i><sub><i>H</i></sub> is 46.1% and 41.1% respectively, which significantly dominates the in-situ stress distribution. This study innovatively combines swarm intelligence optimization with a fuzzy inference system, and achieves dual breakthroughs in high-precision prediction and mechanism interpretation through the PSO-ANFIS-SHAP framework, providing a reliable method for in-situ stress field inversion and reservoir stimulation evaluation in areas with low exploration degree.</p>

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

ANFIS-PSO based in-situ stress prediction model for coal reservoirs: a case study in Wenjiaba Block, Guizhou

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月14日

查看全文
  • Zhongman Luo,
  • Chao Pan,
  • Xuefeng Li,
  • Haiying Ren,
  • Juncai Cao,
  • Meilu Yu,
  • Jilin Li,
  • Furu Kang

摘要

Accurate prediction of the ground stress field is an important scientific issue in the evaluation of the compressibility of coalbed methane reservoirs and the design of hydraulic fracturing. This study addresses the challenges of multi-source data fusion and insufficient generalization capability for small samples in predicting in-situ stress in complex coal reservoirs. It proposes an in-situ stress prediction model ANFIS-PSO that integrates the Particle Swarm Optimization (PSO) algorithm with an Adaptive Neuro-Fuzzy Inference System (ANFIS). The model integrates multi-source data including acoustic time-of-flight logging, measured in-situ stress values from injection/depression tests, and Brazilian splitting and uniaxial compression experiments. Through Pearson correlation analysis (|R|≥ 0.31, p < 0.001) to identify five key input parameters: GR, LLS, AC, DEN, and TVD. This established a nonlinear mapping relationship between logging responses and in-situ stress. The PSO algorithm was employed to globally optimize the Gaussian membership function parameters and fuzzy rule weights of the ANFIS model. After 228 iterations, the model's convergence and accuracy were significantly enhanced. The results show that: compared with the MLP-PSO and RF-PSO, the RMSE of the ANFIS-PSO for predicting the minimum horizontal principal stress (σh) decreases by 25.0% and 39.2% respectively; for predicting the maximum horizontal principal stress (σH), the RMSE decreases by 31.8% and 72.7% respectively, and the coefficient of determination R2 ranges from 0.96 to 0.97 (n = 216). The SHAP interpretability analysis indicates that the contribution of true vertical depth (TVD) to σh and σH is 46.1% and 41.1% respectively, which significantly dominates the in-situ stress distribution. This study innovatively combines swarm intelligence optimization with a fuzzy inference system, and achieves dual breakthroughs in high-precision prediction and mechanism interpretation through the PSO-ANFIS-SHAP framework, providing a reliable method for in-situ stress field inversion and reservoir stimulation evaluation in areas with low exploration degree.