基于生成对抗网络的半监督地震波阻抗反演

doi:10.19657/j.geoscience.1000-8527.2023.089

摘要/Abstract

摘要：

深度学习波阻抗反演通常需要大量的标记数据对网络进行训练,然而在实际应用中标记数据(测井数据)往往是有限的,针对此问题,本文提出了一种基于生成对抗网络(GAN)的半监督地震反演新方法。该方法采用有条件的GAN(cGAN)改进传统GAN网络结构,重新设计U型卷积神经网络(Unet)的生成器和残差网络(Resnet)的判别器,并采用Wasserstein GAN (WGAN)构建新的目标函数。网络训练分两个阶段,先用少量标记数据训练判别器,再用少量标记数据和大量未标记数据训练生成器,其中生成器受到正演褶积模型约束。合成数据实验结果表明,本文提出的方法适用于少量标记数据的波阻抗反演问题,可以准确反演出波阻抗模型,且具有较好的抗噪性能;实测资料反演结果表明本方法具有较好的实用性。该方法对解决地震波阻抗反演中标记数据少的问题提供了新的参考方法,具有较好的实际应用前景。

关键词: 深度学习, 地震波阻抗反演, 生成对抗网络, 半监督学习

Abstract:

Deep learning impedance inversion usually requires a large amount of labeled data for network training. However, in practical applications, labeled data (well logging data) is often limited. To address this issue, this paper proposes a new semi-supervised seismic inversion method based on Generative Adversarial Networks (GANs). The method improves the traditional GAN network structure by using a conditional GAN (cGAN), redesigning the generator with a Unet structure and the discriminator with a Resnet structure, and employing Wasserstein GAN (WGAN) to construct a new objective function. The network training is divided into two stages: first, training the discriminator with a small amount of labeled data, then, training the generator with a small amount of labeled data and a large amount of unlabeled data, where the generator is constrained by the forward convolutional model. Experimental results on synthetic data demonstrate that the proposed method is suitable for impedance inversion with limited labeled data, accurately reconstructs impedance models, and exhibits good noise resistance. Inversion results on real field data also indicate the practicality of this method. This method offers a novel approach for addressing the challenge of limited labeled data in seismic wave impedance inversion, holding promising prospects for practical applications.

Key words: deep learning, seismic wave impedance inversion, generative adversarial network, semi-supervised learning

中图分类号:

P631.4

王永昌, 刘彩云, 熊杰, 王康, 胡焕发, 康佳帅. 基于生成对抗网络的半监督地震波阻抗反演[J]. 现代地质, 2024, 38(06): 1585-1593.

WANG Yongchang, LIU Caiyun, XIONG Jie, WANG Kang, HU Huanfa, KANG Jiashuai. Semi-supervised Seismic Wave Impedance Inversion Based on Generative Adversarial Networks[J]. Geoscience, 2024, 38(06): 1585-1593.

图/表 15

图1 地震波阻抗反演流程图

Fig.1 Seismic impedance inversion process

图2 地震波阻抗反演判别器结构 (a)残差块结构;(b)判别器结构

Fig.2 Discriminator structure for seismic impedance inversion

图3 地震波阻抗反演生成器结构

Fig.3 Structure of seismic impedance inversion generator

图4 地震波阻抗反演合成地震数据

Fig.4 Seismic impedance inversion to synthesize seismic data

图5 地震波阻抗反演结果及对比 (a)真实地震波阻抗;(b)生成器反演地震波阻抗;(c)地震波阻抗误差;(d)Wu等[19]反演结果

Fig.5 Seismic impedance inversion results and comparison

表1 地震波阻抗反演性能统计

Table 1 Performance statistics of seismic impedance inversion

参数	本文方法	Wu等^[19]方法
PCC	0.9985	0.9948
r²	0.9966	0.9874
MSE	0.0029	0.0184

图6 第2700道的反演结果(a)和第7440道的反演结果(b)

Fig.6 Inversion results of NO.2700(a) and NO.7440(b)

表2 地震波阻抗带噪数据反演性能统计

Table 2 Performance statistics of seismic impedance inversion with noisy data

参数	25db	15db	5db
PCC	0.9979	0.9902	0.8655
r²	0.9952	0.9783	0.5581
MSE	0.0043	0.0192	0.3253

图7 地震波阻抗加噪反演结果 (a)真实阻抗;(b)SNR为25 db的反演结果;(c)SNR为15 db的反演结果;(d)SNR为5 db的反演结果

Fig.7 Seismic impedance with noisy inversion results

图8 地震波阻抗SEAM模型

Fig.8 Seismic impedance SEAM model

图9 地震波阻抗SEAM反演结果 (a)本文方法反演结果;(b)Wu等[19]反演结果

Fig.9 Seismic impedance SEAM inversion results

表3 地震波阻抗SEAM模型反演性能统计

Table 3 Performance statistics of seismic impedance inversion for SEAM model

参数	本文方法	Wu等^[19]方法
PCC	0.9961	0.9902
r²	0.9911	0.9753

图10 第750道和第800道反演结果 (a)(b)本文方法结果;(c)(d)Wu等[19]方法结果

Fig.10 Inversion result of NO.750 and inversion result of NO. 800

图11 Volve实测地震数据[17] (a)北海近海Volve油田;(b)地震数据和井轨迹

Fig.11 Seismic data from the Volve field[17]

图12 Volve反演结果

Fig.12 Volve inversion results

参考文献 25

[1]	BACKUS G, GILBERT F. Uniqueness in the inversion of inaccurate gross Earth data[J]. Philosophical Transactions of the Royal Society of London (Series A: Mathematical & physical sciences), 1970, 266: 123-192.
[2]	沙志彬, 郑涛, 杨木壮, 等. 基于波阻抗反演的天然气水合物地震检测技术[J]. 现代地质, 2010, 24(3): 481-488.
[3]	李芳, 邓勇, 胡林, 等. 地球物理技术预测莺歌海盆地低孔低渗储层孔隙度[J]. 地质科技通报, 2022, 41(4):84-90.
[4]	SUN M N, JIN S G. Multiparameter elastic full waveform inversion of ocean bottom seismic four-component data based on a modified acoustic-elastic coupled equation[J]. Remote Sensing, 2020, 12(17): 2816.
[5]	YANG H, JIA J X, WU B Y, et al. Mini-batch optimized full waveform inversion with geological constrained gradient filtering[J]. Journal of Applied Geophysics, 2018, 152: 9-16.
[6]	MADIBA G B, MCMECHAN G A. Seismic impedance inversion and interpretation of a gas carbonate reservoir in the Alberta Foothills, Western Canada[J]. Geophysics, 2003, 68(5): 1460-1469.
[7]	LECUN Y, BENGIO Y, HINTON G. Deep learning[J]. Nature, 2015, 521: 436-444.
[8]	XIONG W, JI X, MA Y, et al. Seismic fault detection with convolutional neural network[J]. Geophysics, 2018, 83(5):97-103. DOI
[9]	王钰清, 陆文凯, 刘金林, 等. 基于数据增广和CNN的地震随机噪声压制[J]. 地球物理学报, 2019, 62(1): 421-433. DOI
[10]	WANG Y Q, WANG Q, LU W K, et al. Seismic impedance inversion based on cycle-consistent generative adversarial network[J]. Petroleum Science, 2022, 19(1): 147-161.
[11]	ZHANG Z P, LIN Y Z. Data-driven seismic waveform inversion: A study on the robustness and generalization[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(10): 6900-6913.
[12]	WU B Y, MENG D L, WANG L L, et al. Seismic impedance inversion using fully convolutional residual network and transfer learning[J]. IEEE Geoscience and Remote Sensing Letters, 2020, 17(12): 2140-2144.
[13]	WANG Y Q, GE Q, LU W K, et al. Well-logging constrained seismic inversion based on closed-loop convolutional neural network[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(8): 5564-5574.
[14]	PAN W, TORRES-VERDÍN C, PYRCZ M J. Stochastic Pix2pix: A new machine learning method for geophysical and well conditioning of rule-based channel reservoir models[J]. Natural Resources Research, 2021, 30(2): 1319-1345.
[15]	LI S C, LIU B, REN Y X, et al. Deep-learning inversion of seismic data[J]. IEEE Transactions on Geoscience and Remote Sensing, 2020, 58(3): 2135-2149.
[16]	WANG B F, ZHANG N, LU W K, et al. Deep-learning-based seismic data interpolation: A preliminary result[J]. Geophysics, 2019, 84(1): V11-V20. DOI
[17]	DAS V, POLLACK A, WOLLNER U, et al. Convolutional neural network for seismic impedance inversion[J]. Geophysics, 2019, 84(6): 869-880.
[18]	HUANG L, POLANCO M, CLEE T E. Initial experiments on improving seismic data inversion with deep learning[M]//2018 New York Scientific Data Summit (NYSDS). New York: IEEE, 2018: 1-3.
[19]	WU B Y, MENG D L, ZHAO H X. Semi-supervised learning for seismic impedance inversion using generative adversarial networks[J]. Remote Sensing, 2021, 13(5): 909.
[20]	GOODFELLOW I J, POUGET-ABADIE J, MIRZA M, et al. Generative adversarial networks[EB/OL]. ArXiv, 2014: 1406.2661. http://arxiv.org/abs/1406.2661.
[21]	MIRZA M, OSINDERO S. Conditional generative adversarial nets[EB/OL]. ArXiv, 2014: 1411.1784. http://arxiv.org/abs/1411.1784.
[22]	ARJOVSKY M, CHINTALA S, BOTTOU L. Wasserstein Gan[EB/OL]. ArXiv, 2017:1701.07875. http://arxiv.org/abs/1701.07875.
[23]	GULRAJANI I, AHMED F, ARJOVSKY M, et al. Improved training of Wasserstein GANs[EB/OL]. ArXiv, 2017: 1704.00028. http://arxiv.org/abs/1704.00028.
[24]	MARTIN G S, WILEY R, MARFURT K J. Marmousi2: An elastic upgrade for Marmousi[J]. The Leading Edge, 2006, 25(2): 156-166.
[25]	SONG L, YIN X Y, ZONG Z Y, et al. Semi-supervised learning seismic inversion based on spatio-temporal sequence residual modeling neural network[J]. Journal of Petroleum Science and Engineering, 2022, 208: 109549.