井地波场特征联合的碳酸盐岩溶蚀孔缝型储层智能识别

深层溶蚀孔缝型储层的精细识别是油气勘探领域重点关注的前沿问题之一，单独采用地面地震或VSP资料开展识别均存在局限性。为充分结合VSP的精细刻画和地面地震广阔照明优势，本项研究围绕溶蚀缝洞储层的VSP与地面地震波场响应高维特征提取、样本不确定条件下碳酸盐岩溶蚀缝洞储层智能识别这两项关键科学问题开展研究，采用VSP与地面地震结合、先验知识与机器学习融合、正演模拟与实际资料分析、多学科交叉的研究思路进行创新研究。利用VSP资料高信噪比、高分辨率、多分量特性，采用高维信号模式分析获得高可靠性的溶蚀孔缝储层VSP识别结果；以VSP识别结果为样本，通过研究样本不确定条件下的智能识别算法，实现大面积的溶蚀孔缝储层井地联合智能识别。主要创新点包括：先验知识引导的溶蚀孔缝储层高维波场深度特征提取方法、基于VSP高维信号反射模式分析的储层识别和样本生成方法、样本不确定性条件下井地波场特征联合储层智能识别方法。

Deep carbonate reservoirs dominated by a vugular-solution porosity system bring significant challenges to seismic exploration and thus, make their seismic characterization a cutting-edge topic in both industry and academia. Neither the surface seismic nor the VSP（vertical seismic profile） survey alone could offer a satisfactory solution. VSPs, benefited from the high-resolution multi-component recordings which relate closely to the fracture network as well as saturants, show advantages over the surface seismic surveys in terms of characterizing the fractured-vuggy network as well as the saturants. Seismic surveys, on the other hand, although suffer from their degraded data quality, have wide illimitation ranges compared with VSPs who only cover the vicinity of the borehole. To combine the strength of both surveys, this project involves multidisciplinary techniques and subjects in signal processing. Relying on the multi-component reflection pattern analysis of the VSP data, this project　intents to make full use of the both P and S-component such that the reservoir characterization is refined to the extent of multi-wavemodes at their highest resolution possible. The prior-constrains obtained from VSP are integrated with the traditional deep neural network (DNN) and thus, a novel informed-network is established to predict the fractured-vuggy carbonate reservoirs from surface seismic. While the integrity of reservoir labels interpreted from VSP varies, this project addresses such uncertainties by introducing a measure of the label fidelity in the deep learning process. After all, this project brings a novel joint VSP-surface seismic method engined by an informed-DNN for an enhanced fractured-vuggy carbonate reservoirs characterization. Three innovative achievements are proposed: 1) A novel prior-information driven deep feature extraction method for the multi-dimension/mode wavefield; 2) An enhanced VSP reservoir characterization driven by multi-component reflection pattern analysis; 3) A DNN based joint VSP-surface seismic method accommodating the uncertainties in label fidelity for the fractured-vuggy carbonate reservoirs characterization.