测井约束高维闭环网络地震波阻抗反演数据集

公开了属于地球物理技术领域的一种测井约束高维闭环网络地震波阻抗数据。三维叠后地震数据，Inline方向共680道地震道，Crossline方向共767道地震道，每道400-450个采样点，采样频率1000HZ。整个地震区共4口测井。使用该数据，具体包括如下步骤：步骤1：搭建卷积神经网络；包括一个正演网络和一个反演网络；步骤2：准备训练数据；包括测井波阻抗数据、插值波阻抗数据和合成地震数据；步骤3：训练网络并进行微调：对正演网络和反演网络进行训练，然后将一维的测井数据应用到二维模型和三维模型上，并进行微调；步骤4：预测及评估：首先对地震数据进行反演；然后对反演结果进行正演得到重构地震数据；最后使用该重构地震数据对反演结果的有效性进行评估。本方法无需额外收集输入和参考图像，能够保证良好的横向连续性，精度高于传统及其他深度学习反演方法。

A well-logging constrained high-dimensional closed-loop network seismic wave impedance dataset in the field of geophysical technology is publicly released. The dataset includes 3D post-stack seismic data with 680 seismic traces in the Inline direction and 767 seismic traces in the Crossline direction; each trace contains 400 to 450 sampling points, with a sampling frequency of 1000 Hz. A total of 4 well logs are provided for the entire seismic survey area.

Specific workflows for utilizing this dataset are outlined below:
1. Construct a convolutional neural network (CNN) comprising a forward modeling network and an inversion network;
2. Prepare training data, which consists of well-logging wave impedance data, interpolated wave impedance data, and synthetic seismic data;
3. Train and fine-tune the networks: First train the forward modeling and inversion networks, then apply the one-dimensional well-logging data to both 2D and 3D models for additional fine-tuning;
4. Perform prediction and evaluation: First conduct seismic inversion on the input seismic data, then perform forward modeling on the inversion results to generate reconstructed seismic data, and finally use the reconstructed seismic data to assess the effectiveness of the inversion outcomes.

This method eliminates the need for additional collection of input and reference seismic images, ensures excellent lateral continuity, and achieves higher accuracy than traditional and other deep learning inversion methods.

搜集汇总

背景与挑战

背景概述

该数据集属于地球物理技术领域，提供了测井约束的高维闭环网络地震波阻抗反演数据，包括三维叠后地震数据和测井信息。它采用卷积神经网络方法进行反演和正演评估，无需额外输入，具有较高的横向连续性和精度优势。