长庆油田某区块地层岩石物理参数智能表征数据集

为解决测井数据在油气勘探开发中的缺失与质量不一问题，构建了一套基于岩石物理响应机制的测井数据正演模拟与智能生成体系，结合深度学习与地质领域知识，提出了高质量、多样化测井数据集的构建方法。首先，通过建立综合性的岩石物理正反演模型，模拟常规与特殊测井曲线响应特征，进而基于Encoder-Decoder结构设计神经网络，实现地质参数向测井曲线的高精度映射。通过引入测井响应方程作为物理约束，并叠加残差网络提升非线性关系建模能力，大幅提升了生成数据的物理合理性与地质一致性。此外，研究还提出了无监督学习的数据集扩充策略，通过分析测井曲线与核磁T2谱的关系，有效填补了特殊测井数据的缺口。同时，为支撑测井储层参数预测任务，构建了系统的数据标签体系，涵盖回归与分类多任务输出，实现了机理解释、岩心分析与试油结论三类标签的有机融合。最终在鄂尔多斯盆地姬塬地区构建了高质量结构化数据集，覆盖4000余口井，数据经过标准化清洗、归一化处理与标签一致性校验，具备良好的建模适配性。

To address the issues of missing data and inconsistent quality of well logging data in oil and gas exploration and development, we developed a well logging data forward simulation and intelligent generation system based on rock physical response mechanisms, integrating deep learning and geological domain expertise, and proposed a methodology for constructing high-quality and diverse well logging datasets. First, we established a comprehensive rock physical forward and inverse model to simulate the response characteristics of conventional and special well logging curves. Subsequently, we designed a neural network based on the Encoder-Decoder architecture to achieve high-precision mapping from geological parameters to well logging curves. By introducing the well logging response equation as a physical constraint and incorporating Residual Networks (ResNet) to enhance the modeling capability of nonlinear relationships, we significantly improved the physical rationality and geological consistency of the generated data. In addition, this study proposed an unsupervised learning-based dataset augmentation strategy, which effectively fills the gaps in special well logging data by analyzing the relationship between well logging curves and nuclear magnetic resonance (NMR) T2 spectra. Meanwhile, to support the well logging reservoir parameter prediction task, we built a systematic data label system that covers multi-task outputs for regression and classification, achieving the organic integration of three types of labels: mechanistic interpretation, core analysis, and well test conclusions. Finally, we constructed a high-quality structured dataset in the Jiyuan Area of the Ordos Basin, covering more than 4000 wells. The data has undergone standardized cleaning, normalization processing, and label consistency verification, exhibiting excellent modeling adaptability.

搜集汇总

背景与挑战

背景概述

该数据集为解决油气勘探开发中测井数据缺失与质量不一问题，构建了一套基于岩石物理响应机制的测井数据正演模拟与智能生成体系，结合深度学习与地质领域知识，提出了高质量、多样化测井数据集的构建方法。数据集覆盖鄂尔多斯盆地姬塬地区4000余口井，数据量达3.63GB，包含155个文件，经过标准化清洗、归一化处理与标签一致性校验，具备良好的建模适配性。