Abstract:

Natural fractures strongly affect flow capacity and development performance in tight sandstone reservoirs. In the Chang 8 Member of the southwestern Yishan Slope, Ordos Basin, thin sandstone-mudstone interbeds, strong fracture heterogeneity, fracture filling, and borehole conditions complicate conventional logging responses, making single-log thresholds unsuitable for continuous fracture identification. Based on core observations, interpretations of electrical imaging logs, and conventional well logs, this study analyzed the development characteristics and logging-response differences of natural fractures in the Chang 8 Member and developed a Neighborhood Reference-Constrained LogFormer (NRC-LogFormer) model for fracture identification. The model uses local neighborhood responses within each well as references and jointly captures center-neighborhood differences, multiscale perturbations, and sequential context, thereby improving the identification of weak fracture responses and reducing misidentification caused by lithologic changes and borehole disturbances. The results show that structural fractures and bedding-parallel/lithologic-interface fractures dominate the study area, with high-angle fractures being prevalent. Fractures with dips of 70°~90° account for 72.4%, and those with dips greater than 60° account for 82.3%. Fractures are more developed in Chang 8₁ than in Chang 8₂ and mainly occur in sandstone and fine sandstone, with their distribution jointly influenced by tectonic stress, sand-body distribution, and lithologic interfaces. Compared with stable non-fracture backgrounds within the same well, intervals containing interpreted fracture points generally exhibit lower sonic transit time, higher compensated density, lower compensated neutron, and differentiated resistivity responses, characterized by lower deep induction resistivity and eight-laterolog resistivity, but higher formation resistivity. These response patterns may be related to host lithology, fracture filling, mud-filtrate invasion, and measurement-volume effects. Caliper, gamma ray, and spontaneous potential provide limited independent discrimination, indicating that fracture-related logging responses are multivariate and non-unique. The model achieved precision, recall, and F₁ values of 0.875, 0.863, and 0.869, respectively, outperforming support vector machine (SVM), extreme gradient boosting (XGBoost), one-dimensional convolutional neural network, and Transformer models. Applications to representative wells showed that high-probability fracture intervals agreed well with electrical imaging-log interpretations and microresistivity circumferential imaging dynamic images. Prediction uncertainty can further assist in identifying suspected fracture responses and intervals requiring focused review. The proposed method provides support for fracture-interval screening, logging-interpretation review, and reservoir evaluation in tight sandstone reservoirs where electrical imaging-log data are limited.

Key words: Ordos Basin, Chang 8 Member, tight sandstone reservoir, natural fractures, logging response characteristics, fracture identification

摘要：

天然裂缝是影响致密砂岩储层渗流能力和开发效果的重要因素，但鄂尔多斯盆地伊陕斜坡西南部长8段砂泥岩薄互层发育，裂缝非均质性强，常规测井响应受岩性变化、裂缝充填和井眼条件共同影响，依靠单曲线阈值难以实现裂缝段连续识别。本文以岩心观察、电成像测井解释和常规测井资料为基础，分析长8段天然裂缝发育特征及测井响应差异，建立邻域参照约束测井序列模型(NRC-LogFormer)。该模型以同井局部邻域响应为参照，融合中心与邻域的响应差异、多尺度扰动特征及序列上下文信息，可增强弱裂缝响应识别并减少岩性突变和井眼扰动造成的误判。结果表明，研究区天然裂缝以构造裂缝和层理缝/岩性界面缝为主，高角度裂缝占优，其中，70°~90°裂缝占72.4%，大于60°裂缝占82.3%；长8₁裂缝发育强于长8₂。裂缝主要赋存于砂岩和细砂岩中，受构造应力、砂体展布和岩性界面共同影响。与同井稳定非裂缝背景相比，裂缝解释点所在层段总体呈低声波时差、高补偿密度、低补偿中子及差异化电阻率响应，其中，深感应电阻率和八侧向电阻率相对降低，地层电阻率相对升高。该组合可能与裂缝赋存岩性、充填状态、泥浆侵入和测井体积效应有关，井径、自然伽马和自然电位的独立指示能力较弱，说明长8段裂缝测井响应具有明显组合性和非唯一性。验证结果显示，模型精确率、召回率和F₁值分别为0.875、0.863和0.869，均优于支持向量机、极端梯度提升、一维卷积神经网络和Transformer模型。典型井应用表明，高概率裂缝段与电成像裂缝解释及井周微电阻率成像测井动态图像对应较好，预测不确定性可辅助识别疑似裂缝响应和需重点复核井段。该方法可为电成像资料有限条件下致密砂岩储层裂缝段筛选、测井解释复核和储层评价提供参考。

关键词: 鄂尔多斯盆地, 长8段, 致密砂岩储层, 天然裂缝, 测井响应特征, 裂缝识别

CLC Number: 

• P618.13

• TE122