西安石油大学学报（自然科学版）2026，Vol.41Issue(3)：55-64,10.DOI:10.3969/j.issn.1673-064X.2026.03.006
基于AutoML-SHAP可解释模型的深层页岩可压性评价方法
Deep Shale Fracability Evaluation Method Based on AutoML-SHAP Interpretable Model
摘要
Abstract
Hydraulic fracturing is a core technology for enhancing the permeability and conductivity of unconventional reservoirs.Accurate e-valuation of reservoir Fracability before fracturing is crucial for efficient development of oil and gas.To solve the problems of generalization a-bility relying on feature engineering and insufficient interpretability of conventional artificial intelligence evaluation methods,a new method for evaluation of reservoir Fracability combining data-driven and mechanism constraint is proposed.Taking the Longmaxi Formation shale in the Luzhou block of the Sichuan Basin as the object,a geological-engineering sweet spot prediction model was established through AutoML,and the complex nonlinear model was posterior interpreted using SHAP algorithm to reveal the influence mechanism of characteristic parameters on the fracture network;This influence is transformed into physical constraints and a linear reservoir Fracability evaluation model is established.Research has shown that:(1)SHAP analysis quantitatively reveals that porosity and horizontal stress difference are the absolute main control-ling factors of geological and engineering sweet spots,respectively,elucidating the contribution of the joint effect of the main and covariates to reservoir fracability.(2)The physical simulation results are highly consistent with the model prediction results,and the experimentally ob-served change of crack from simple main crack to complex crack network is in good agreement with the engineering sweet spot calculation re-sult.Two sets of real triaxial hydraulic fracturing physical simulation experiments were conducted to verify the rationality of engineering sweet spot differences,and this algorithm was applied to two wells in the Luzhou block to verify that the model has both high prediction accuracy and interpretability,which can provide reference for on-site design.关键词
深层页岩/可压性评价/可解释人工智能/SHAP/物理约束Key words
deep shale/reservoir fracability evaluation/explainable artificial intelligence/SHAP/physical constraint分类
能源科技引用本文复制引用
刘珊,侯冰,谢锦阳,黄毅..基于AutoML-SHAP可解释模型的深层页岩可压性评价方法[J].西安石油大学学报（自然科学版）,2026,41(3):55-64,10.基金项目
新疆维吾尔自治区科技计划项目"新疆高温高压深层钻探井壁失稳机理与控制关键技术研究"(2024B01014) （2024B01014）
国家自然科学基金重点项目"提高超深大斜度井压裂效率的关键力学问题研究"(52334001) （52334001）
