Abstract:

Shallow-water delta front sand bodies are widely developed and represent important hydrocarbon reservoirs with high exploration potential in continental basins. However,frequent migration and avulsion of distributary channels result in vertically stacked thin interbeds and laterally interwoven distributions,leading to complex architectural patterns. Under sparse well coverage,this complexity makes reservoir prediction highly challenging and less accurate. This study takes the Member 1 of Jurassic Shaximiao Formation in the Tianfu Gas Field,Sichuan Basin,as a case example,and proposes an intelligent prediction method for thin sand bodies by integrating multi-frequency seismic attribute automated machine learning(AutoML)with waveform indication inversion(WII). Laterally,optimized frequency-decomposed seismic data were used to automatically select key attributes and integrate multi-model learning,building nonlinear mappings that improved the correlation between seismic attributes and well-derived sand thickness from 0.64 to 0.82. This enabled quantitative prediction of sand bodies under sparse well control. Vertically,WII introduced facies-controlled constraints and used uranium-free gamma ray as the inversion target,allowing stable identification of sand bodies thicker than 10 m and reliable responses from thin layers of about 4 m,and establishing seismic response templates of architectural element assemblages. Based on these results,four types of architectural elements were identified: distributary channel,residual mouth bar,levee,and interdistributary bay. Furthermore,six stacking patterns including vertical superimposition and lateral accretion were summarized. The results also reveal that,under different depositional cycles,isolated narrow channels and compound wide channels show distinct spatial distributions and vertical stacking characteristics. In three dimensions,a spatial differentiation pattern was clarified,evolving from near-source net-like distributary channels to distal tongue-shaped and lobe-shaped mouth bars. Finally,a depositional architectural model of the shallow-water delta was established. This study achieves fine-scale multi-level architectural characterization under sparse well conditions and provides new insights and methods for efficient exploration and development of complex shallow-water delta reservoirs.

Key words: shallow-water delta, multi-scale architectural characterization, thin sandstone body prediction, automated machine learning, multi-frequency intelligent attribute fusion, Shaximiao Formation, Sichuan Basin

摘要：

浅水三角洲前缘砂体广泛发育,是陆相盆地中极具勘探潜力的重要油气储集层类型,但受分流河道频繁迁移改道影响,砂体在垂向上多期叠置形成薄互层,平面上相互交织,构型分布较为复杂,致使在井网稀疏条件下储层预测难度大、精度低。本研究以四川盆地天府气田简阳区块侏罗系沙溪庙组一段浅水三角洲为例,提出“分频多属性自动化机器学习联合波形指示反演”的薄层砂体智能预测方法: 在平面上,优选分频地震数据,自动筛选关键属性并集成多模型学习,建立非线性映射关系,使地震属性与井上砂厚相关系数由0.64提升至0.82,实现稀井网条件下砂体的定量刻画; 在剖面上,引入波形指示反演的相控约束,并以无铀伽马为目标开展反演,稳定识别10 m以上砂体,对约4 m的薄层砂体亦有响应,建立了构型单元组合样式的地震响应模板。基于上述成果,系统识别出水下分流河道、残余河口坝、溢岸及分流间湾4类构型单元,总结形成6种垂向叠置和侧向切叠组合样式,并揭示了在沉积旋回控制下孤立窄河道与复合宽河道在平面展布与垂向差异演化特征,同时在三维上厘清了由近源网状分流河道到远源舌状-朵状坝体的空间分异规律,建立了浅水三角洲沉积构型模式。研究成果不仅实现了稀疏井网下多级次构型的精细表征,也为复杂浅水三角洲储集层的高效勘探开发提供了新思路和方法。

关键词: 浅水三角洲, 多级次构型表征, 薄层砂体预测, 自动化机器学习, 分频智能属性融合, 沙溪庙组, 四川盆地

CLC Number: 

• P539.2