﻿ 训练图像对多点地质统计反演效果的影响

# 训练图像对多点地质统计反演效果的影响The Influence of Training Images on the Effect of Multipoint Geostatistical Inversion

Abstract: The stochastic modeling is developed from two point geostatistics to multi-point geostatistics, and a seismic inversion method based on multi-point geostatistics is proposed. Since the training image is the key of multi-point geostatistical modeling, it directly determines the quality of the modeling results. An evaluation of training image in inversion is necessary. Three different training image is designed to reveal the influence on inversion result, that is, a training image same to the real reservoir, a training image reflecting the structure of the real reservoir, and a rotation of 90 degree which is different to the real reservoir. The results show that the training image has a great influence on the convergence speed of the multi-point geostatistical inversion, and the more ac-curate the training image is, the faster convergence speed of the multi-point geostatistical inversion is. The place of the lithofacies has little influence unless they have different structure.

1. 前言

2. 多点地质统计学反演原理

2.1. 参数确定

1) 岩石弹性参数统计分析

2) 训练图像建立

2.2. 随机反演

3. 模型测试

Figure 1. A 2-dimentional mudstone and sandstone model

Figure 2. The distribution of elastic parameters

Figure 3. The record of seismic data

Figure 4. Training images (a) original training image, (b) the place variation of original training image, (c) the rotation of original training image

Figure 5. A comparison of the first iteration of seismic inversion (a) the result using the original training image, (b) the result using the variable place of training image, (c) the result using the rotation training image

Figure 6. A comparison of after 10th iteration of seismic inversion (a) the result using the original training image, (b) the result using the variable place of training image, (c) the result using the rotation training image

4. 结论

1) 训练图像对多点地质统计反演效果有很大的影响，建立准确的训练图像能够减少反演过程中的迭代次数和反演的计算量。在多点地质统计反演中，地质工作者应该尽量建立准确的训练图像。

2) 训练图像结构类似情况下，砂泥岩位置分布差异对反演结构影响不大。因此在建立训练图像时候，其核心在于揭示地下储层结构信息。

[1] 吴胜和, 李文克. 多点地质统计学——理论/应用与展望.古地理学报[J]. 2005, 7(1).

[2] González, F., Mukerji, T. and Mavko, G. (2008) Seismic Inversion Combining Rockphysics and Multiple-Point Geostatistics. Geophysics, 73, 11-21.
https://doi.org/10.1190/1.2803748

[3] Zhou, H. and Gomez-Hernandez, J. (2012) A Pattern-Search-Based Inverse Method. Water Resources Research, 48.

[4] Zhang, T.f., Switzer, P., Journel, A., et al. (2006) Filter-Based Classification of Training Image Patterns for Spatial Simulation. Mathematical Geology, 38.
https://doi.org/10.1007/s11004-005-9004-x

[5] Larsen, A., Ulv-moen, M., Omre, H., et al. (2006) Bayesian Lithology/Fluid Prediction and Simulation on the Basis of a Markov-chain Prior Model. Geophysics, 71, 69-78.
https://doi.org/10.1190/1.2245469

[6] Tarantola, A. (2005) Inverse Problem Theory and Methods for Model Parameter Estimation. Society for Industrial and Applied Mathematics.
https://doi.org/10.1137/1.9780898717921

[7] Nunes, R., Soares, A., Azevedo, L., et al. (2017) Geostatistical Seismic Inversion with Direct Sequential Simulation and Co-simulation with Multi-local Distribution Functions. Mathematical Geosciences, 49, 583-601.
https://doi.org/10.1007/s11004-016-9651-0

[8] 杨培杰. 地质统计学反演——从两点到多点[J]. 地球物理学进展, 2014, 29(5): 2293-2300.

Top