【作者】 邓光校；

【导师】 宋先海； 顾汉明；

【作者基本信息】 中国地质大学 ， 地球探测与信息技术， 2024， 博士

【摘要】 塔河油田已开发二十多年,其主要开发目的层奥陶系碳酸盐岩溶洞体或缝洞组合的等效溶洞体是油气重要的储集空间,但缝洞储层非均质性严重,地震响应特征复杂,随着开发的深入,传统的溶洞体或缝洞体地震表征方法已满足不了开发阶段井位设计的需要,为了提高油藏精细描述刻画溶洞单元的展布形态的精度,论文开展碳酸盐岩溶洞三维地震响应特征及储层地震定量表征方法研究,具有重要的科学意义和应用价值。论文针对碳酸盐岩溶洞或缝洞组合的储层横向不均匀性强,地震波场特征复杂,地震可识别性不明确及地震预测多解性严重问题,开展了基于不同类型不同形态3D溶洞模型的建立和正演模拟的地震可识别性研究;同时利用不同规模、不同形态的溶洞体三维地震偏移成像数据体及其地震属性构成的样本集,基于模式匹配识别及深度学习的三维溶洞体地震表征方法研究,通过地震合成数据和实际地震数据试算验证了上述方法的可靠性和适用性。论文取得的主要成果和认识如下:(1)采用保持地层倾角不变,纵横向距离缩小的原则可以建立既能将大范围工区的地层格架模型缩小到正演模拟所要求的模型大小又能保持格架地层构造特点的模型,基于采用人机交互可以建立具有规则形状的不同空间尺度和分布特点的溶洞组合模型,基于实际缝洞地震数据属性异常的刻画和空间体积校正,可以建立接近实际缝洞发育规模的不同大小不同形态的等效溶洞模型,为三维溶洞地震波场可识别性及溶洞地震定量表征提供了较完备的模型集。(2)采用双重网格大小的交错网格GPU和CPU协同并行三维波动方程正演模拟方法,不仅在空间上有效提高塔河油田奥陶系不同尺度溶洞三维波动方程正演模拟的精度,而且降低计算模型所需内存消耗和计算耗时,提高了计算效率,为三维溶洞地震波场可识别性及溶洞地震定量表征提供了地震数据集。(3)基于3D溶洞模型对应的叠前深度偏移成像剖面及水平切片,同时,基于地震属性分析技术获得的不同规模溶洞对应地震属性值的变化特征,可以有效分析溶洞地震波场的可检测性。(4)提出了基于模式识别的三维溶洞体地震表征方法,充分利用了三维溶洞体的地震偏移成像数据体包含溶洞体的个性特征,基于特征提取算法提取分别表征溶洞体规模、溶洞体外部形态以及溶洞体内部充填速度的多种地震属性体,建立三维溶洞模型正演地震属性的样本集;利用三维缝洞模型和地震属性数据集进行缝洞模式识别和匹配,可较好地应用于工区实际地震资料进行不同类型的溶洞的识别。(5)提出了基于深度学习的三维溶洞体地震定量表征方法,该方法基于三维孔洞体模型的地震偏移成像数据体多域样本集建立基础上,划分训练集、测试集和验证集,通过U-Net训练,获得深度学习的溶洞体地震识别的网络结构和超参数模型,应用于实际地震数据溶洞体定量识别,取得了较好的应用效果。本论文的主要贡献及创新之处主要表现在以下三个方面:(1)提出了基于模式识别的三维溶洞体地震表征方法,该方法充分利用了三维溶洞体的地震偏移成像数据体包含溶洞体的个性特征,基于三维溶洞模型和地震属性数据集进行溶洞模式识别和匹配,较好地应用于工区实际地震资料进行不同类型的溶洞的识别。(2)提出了基于深度学习的三维溶洞体地震定量表征方法,该方法基于三维溶洞体模型的地震偏移成像数据体样本集建立基础上,通过U-Net训练,获得深度学习的溶洞体地震识别的网络结构和超参数模型,在塔河实际地震数据溶洞体定量识别中取得了较好的应用效果。更多还原

【Abstract】 The Tahe Oilfield has been developed for over twenty years.The main target reservoirs for development are equivalent karst cave bodies or fracture-cave combinations in the Ordovician carbonate formations,which serve as crucial storage spaces for oil and gas.However,the heterogeneity of fracture-cave reservoirs is significant,and their seismic response characteristics are complex.With the deepening of development,traditional seismic characterization methods for karst cave bodies or fracture-cave bodies no longer meet the needs of well positioning in the development stage.In order to enhance the accuracy of fine reservoir description and depict the distribution morphology of cave units,this thesis explores the three-dimensional seismic response characteristics of carbonate karst caves and quantitative seismic characterization methods for reservoirs.This research holds significant scientific importance and practical value.The thesis addresses the issues of strong lateral heterogeneity in reservoirs formed by carbonate karst caves or fracture-cave combinations,complex seismic wavefield characteristics,unclear seismic detectability,and significant ambiguity in seismic prediction.To tackle these challenges,the study conducts research on seismic detectability based on the establishment of 3D karst cave models of different types and morphologies,as well as forward simulation.Additionally,it explores seismic characterization methods for 3D karst cave bodies using pattern recognition and deep learning,leveraging 3D seismic migration imaging datasets and seismic attributes derived from karst caves of various scales and morphologies.The reliability and applicability of these methods are validated through tests using synthetic seismic data and actual seismic data.The main achievements and insights obtained from the thesis are as follows:(1)Using the principle of maintaining the unchanged dip angle of the strata while reducing the longitudinal and transverse distances can establish a model framework that not only reduces the size of the stratum framework model for a large work area to the size required for forward simulation but also maintains the structural characteristics of the framework strata.Based on human-computer interaction,models with regular shapes and different spatial scales and distribution characteristics of karst cave combinations can be established.Additionally,based on the characterization of anomalous seismic data attributes of actual fracture caves and spatial volume calibration,models of equivalent caves of different sizes and shapes close to the actual development scale of the fracture caves can be established,providing a comprehensive set of models for the recognition of 3D karst cave seismic wavefields and quantitative characterization of karst cave seismicity.(2)Forward modeling method of three-dimensional wave equation using staggered grid GPU and CPU with double grid size not only effectively improves the accuracy of the forward simulation of three-dimensional wave equations for different scales of Ordovician karst caves in the Tahe Oilfield,but also reduces the memory consumption and computational time required by the computational model,thereby enhancing computational efficiency.This method provides seismic data sets for the recognition of3 D karst cave seismic wavefields and quantitative characterization of cave seismicity.(3)Based on the pre-stack depth migration imaging profiles and horizontal slices corresponding to the 3D cave model,as well as the variation characteristics of seismic attribute values corresponding to different scales of caves obtained through seismic attribute analysis techniques,the detectability of cave seismic wavefields can be effectively analyzed.(4)The thesis proposes a pattern recognition-based method for seismic characterization of three-dimensional cave bodies,which fully utilizes the individual characteristics of cave bodies contained in the seismic migration imaging dataset of three-dimensional cave bodies.Based on feature extraction algorithms,various seismic attribute volumes representing the scale,external morphology,and internal fill velocity of cave bodies are extracted to establish a sample set for seismic attribute forward simulation of three-dimensional cave models.By using the three-dimensional karst cave model and seismic attribute dataset for karst cave pattern recognition and matching,this method can be effectively applied to identify different types of caves in actual seismic data within the work area.(5)The thesis proposes a three-dimensional karst cave seismic quantitative characterization method based on deep learning.This method is built upon a multi-domain sample set of seismic migration imaging data of three-dimensional cave models.By partitioning the dataset into training,testing,and validation sets,and utilizing deep residual networks(Res Net)and U-Net for training,the method obtains a deep learning network structure and hyperparameter model for cave seismic identification.This approach is then applied to quantitatively identify caves in actual seismic data,achieving favorable application results.The main contributions and innovations of this thesis are primarily manifested in the following three aspects:(1)The thesis proposes a pattern recognition-based seismic characterization method for three-dimensional cave bodies.This method fully utilizes the individual characteristics of three-dimensional cave bodies contained in seismic migration imaging datasets.By conducting pattern recognition and matching of caves using a three-dimensional cave model and seismic attribute dataset,the method is effectively applied to identify different types of caves in actual seismic data within the work area.(2)The thesis introduces a deep learning-based seismic quantitative characterization method for three-dimensional cave bodies.This method is built upon a sample set of seismic migration imaging data from three-dimensional cave models.By training deep residual networks(Res Net)and U-Net,the method obtains a deep learning network structure and hyperparameter model for cave seismic identification.It has achieved favorable application results in quantitatively identifying caves in actual seismic data from the Tahe Oilfield.更多还原