摘要：

南海涠洲油田L-3段砂泥薄互层大斜度井压裂改造中，射孔靶点选择直接决定裂缝穿层效果和有效改造程度。针对该类储层纹层弱面发育、岩性纵向变化快、层间应力约束复杂，导致优质压裂靶点难以定量识别的问题，本文开展三轴压缩实验、实验前后CT扫描和XRD测试，分析细砂岩、粉砂岩、泥质粉砂岩和泥页岩4类岩性的破坏模式及裂缝扩展特征；进一步融合测井解释岩石力学参数、地应力参数、矿物组成和井眼轨迹数据，构建面向压裂靶点优选的缝高预测特征体系，并在AutoML框架下对GBR、HGBR和MLP模型进行筛选与优化。结果表明，纹层及天然弱结构显著增强岩石非均质性，降低岩石强度，并控制裂缝起裂位置与扩展路径；其中泥质粉砂岩和泥页岩中高密度纹层更易诱导裂缝沿弱面扩展或转向，是造成不同靶点压裂响应差异的关键因素。模型评价显示，MLP在交叉验证阶段取得最低RMSE，为6.37；GBR在独立测试集上表现最优，RMSE和MAE分别为3.71和3.59，表明所建模型可为小样本条件下靶点缝高评价提供定量支撑。W1井应用结果表明，裂缝缝高与层间应力差呈负相关；当层间应力差小于2 MPa、纵横向应力差大于20 MPa时，更易形成缝高超过40 m的穿层裂缝；在连续泥质粉砂岩储层中，高脆性段B1>0.6、B2>80%更有利于获得较大缝高。研究认为，低层间应力差、高纵横向应力差和高脆性是涠洲L-3段砂泥薄互层大斜度井优质压裂靶点的重要判别标志，所建立的缝高预测与靶点评价方法可为海上薄互层储层射孔位置优选、压裂层段排序和施工参数优化提供依据。

关键词: 大斜度井, 砂泥薄互层, 压裂靶点优选, 裂缝缝高, 层间应力差, 纹层弱面, 小样本机器学习

Abstract:

In the fracturing stimulation of highly deviated wells within the sand-mud thin interbeds of the L-3 formation in the Weizhou Oilfield of the South China Sea, the selection of perforation targets directly determines the cross-layer fracture propagation effect and the effective stimulation degree. Such reservoirs are characterized by well-developed laminated weak planes, rapid vertical lithological variation, and complex interlayer stress constraints, which make it difficult to quantitatively identify optimal fracturing targets. To address this problem, triaxial compression tests, pre- and post-experiment CT scanning, and XRD tests were conducted to investigate the failure modes and fracture propagation characteristics of four lithologies: fine sandstone, siltstone, argillaceous siltstone, and mud shale. Furthermore, integrating logging-interpreted rock mechanical parameters, in-situ stress parameters, mineral composition, and well trajectory data, this study constructed a feature system for fracture height prediction oriented to the optimization of fracturing targets. Within the AutoML framework, the GBR, HGBR and MLP models were screened and optimized. The results reveal that laminations and natural weak structures remarkably enhance rock heterogeneity, reduce rock strength, and dominate fracture initiation locations and propagation paths. Among them, high-density laminations developed in argillaceous siltstone and mud shale tend to induce fractures to propagate or deflect along weak planes, acting as the key factor causing differences in fracturing response at various perforation targets. Model evaluation indicates that the MLP model achieves the minimum RMSE of 6.37 in the cross-validation stage, while the GBR model delivers the best performance on the independent test set, with RMSE and MAE of 3.71 and 3.59 respectively. It is verified that the established models can provide quantitative support for the fracture height evaluation of perforation targets under small sample conditions. The application results of Well W1 show that fracture height is negatively correlated with interlayer stress difference. When the interlayer stress difference is less than 2 MPa and the vertical-horizontal stress difference exceeds 20 MPa, cross-layer fractures with a height of more than 40 m are more likely to be formed. In continuous argillaceous siltstone reservoirs, high-brittleness intervals with B1>0.6 and B2>80% are more favorable for acquiring a larger fracture height. This study concludes that low interlayer stress difference, high vertical-horizontal stress difference, and high brittleness are important discrimination indicators for high-quality fracturing targets of highly deviated wells in the L-3 sand-mud thin interbeds of the Weizhou Oilfield. The proposed fracture height prediction and target evaluation method can provide a reliable basis for the optimization of perforation location, the ranking of fracturing intervals, and the optimization of construction parameters in offshore thin interbedded reservoirs.

Key words: highly deviated well, thin sand-mud interbed, fracturing target optimization, fracture height, interlayer stress difference, laminated weak plane, small-sample machine learning

中图分类号: 

• TE257

• TE52