Scholarlink

<p>Accurately predicting the mechanical response of ultra-deep dolomite (burial depth &gt; 6000&#xa0;m) under high in-situ stress and strong heterogeneity is a major challenge for the development of ultra-deep oil and gas resources. Traditional theories and empirical formulas fail to balance prediction accuracy with computational efficiency. This study establishes a framework from high-fidelity simulation to rapid prediction. First, a dual-Weibull distribution heterogeneous model using the combined finite-discrete element method (FDEM) is developed and validated against experimental data to characterize dolomite’s mesoscopic structure. Subsequently, a modified Long Short-Term Memory (mLSTM) deep learning surrogate model is trained on the FDEM dataset to achieve rapid prediction of the complete mechanical process. The study reveals that: (1) The dual-Weibull framework decouples mesoscopic mechanics by independently controlling grain boundary strength (cohesive elements) and grain stiffness heterogeneity (matrix elements), revealing their differentiated effects on peak strength and elastic modulus. (2) Numerical simulations reveal that mechanical behavior is dominated by confining pressure, which governs the transition from microcrack evolution to macroscopic failure patterns. (3) The mLSTM integrates physical initial conditions (e.g., confining pressure, mineral composition) via an initial state encoding mechanism. This suppresses the accumulation of errors in standard LSTMs, significantly improving prediction accuracy. (4) SHapley Additive exPlanations (SHAP) analysis validates the model’s physical reasonableness, as its feature importance rankings match the FDEM sensitivity analysis. Cross-lithology validation confirms strong generalization. This “refined numerical modeling—rapid surrogate prediction” pathway provides an efficient tool for wellbore stability and hydraulic fracturing optimization in ultra-deep drilling.</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

Mechanical Property Prediction of Ultra-Deep Dolomite Based on Dual-Weibull Distribution FDEM and Modified LSTM

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Hongda Li,
  • Yunhu Lu,
  • Yan Jin,
  • Mian Chen,
  • Zhijun Wu

摘要

Accurately predicting the mechanical response of ultra-deep dolomite (burial depth > 6000 m) under high in-situ stress and strong heterogeneity is a major challenge for the development of ultra-deep oil and gas resources. Traditional theories and empirical formulas fail to balance prediction accuracy with computational efficiency. This study establishes a framework from high-fidelity simulation to rapid prediction. First, a dual-Weibull distribution heterogeneous model using the combined finite-discrete element method (FDEM) is developed and validated against experimental data to characterize dolomite’s mesoscopic structure. Subsequently, a modified Long Short-Term Memory (mLSTM) deep learning surrogate model is trained on the FDEM dataset to achieve rapid prediction of the complete mechanical process. The study reveals that: (1) The dual-Weibull framework decouples mesoscopic mechanics by independently controlling grain boundary strength (cohesive elements) and grain stiffness heterogeneity (matrix elements), revealing their differentiated effects on peak strength and elastic modulus. (2) Numerical simulations reveal that mechanical behavior is dominated by confining pressure, which governs the transition from microcrack evolution to macroscopic failure patterns. (3) The mLSTM integrates physical initial conditions (e.g., confining pressure, mineral composition) via an initial state encoding mechanism. This suppresses the accumulation of errors in standard LSTMs, significantly improving prediction accuracy. (4) SHapley Additive exPlanations (SHAP) analysis validates the model’s physical reasonableness, as its feature importance rankings match the FDEM sensitivity analysis. Cross-lithology validation confirms strong generalization. This “refined numerical modeling—rapid surrogate prediction” pathway provides an efficient tool for wellbore stability and hydraulic fracturing optimization in ultra-deep drilling.