基于非线性定向降维的k近邻致密砂岩储层含气性预测方法

本文提出利用全连接人工神经网络(FANN)进行非线性定向降维并结合k近邻方法分类的致密砂岩储层含气性预测方法。k近邻方法能够依据样本间相似性,针对性地选取对应的部分训练样本建立局部模型,但缺乏含气敏感属性的提取能力,并面临“维度灾难”问题。由于样本中的含气性特征虽然是重要特征,但不一定是主要特征。线性降维方法难以准确提取这些特征。我们通过训练一个合理搭建的FANN并输出其中间低维特征实现对训练样本和待预测样本的非线性定向降维。这种做法既能够增加样本的可分性,同时避免了通过样本低维空间中的最大差异实现降维而改变样本固有分布特征的问题。另外,k近邻方法对降维数据进行分类,还等效于用k近邻方法替代FANN中具有线性分类作用的深层结构,有利于白化FANN的黑箱问题。本方法在具体的物理场景中挖掘机器学习算法的物理内涵,提高了智能方法的可解释性。将本方法应用在实际数据中,预测结果显示本方法能够一定程度上挖掘局部波形属性中蕴含的含气敏感信息,实现小范围的致密砂岩储层精确刻画。

硕士研究生教育国际化发展的实践与思考——以中国石油大学(北京)地球物理学院为例

经济全球化与区域经济一体化显著促进了高等教育的国际化。如何培养一批谙熟国际规则,具有良好沟通能力和宽厚基础知识,以及符合科学技术发展技术方向的国际化创新人才,是高校研究生教育迫切需要解决的问题。中国石油大学(北京)为提升石油人才培养的国际化水平,在地球物理学院开设了硕士研究生全英文课程国际班,制订了国际化与本土化相统一的人才培养方案,推进了研究生教育模式国际化改革,建立了海外优质资源与校内海归青年教师相结合的全英文教学团队,显著提高了校内教学与科研环境的国际化水平。

Influence of inaccurate wavelet phase estimation on seismic inversion

On the assumption that the seismic wavelet amplitude spectrum is estimated accurately, a group of wavelets with different phase spectra, regarded as estimated wavelets, are used to implement linear least-squares inversion. During inversion, except for the wavelet phase, all other factors affecting inversion results are not taken into account. The inversion results of a sparse reflectivity model （or blocky impedance model） show that： （1） although the synthetic data using inversion results matches well with the original seismic data, the inverted reflectivity and acoustic impedance are different from that of the real model. （2） the inversion result reliability is dependent on the estimated wavelet Z transform root distribution. When the estimated wavelet Z transform roots only differ from that of the real wavelet near the unit circle, the inverted reflectivity and impedance are usually consistent with the real model; （3） although the synthetic data matches well with the original data and the Cauchy norm （or modified Cauchy norm） with a constant damping parameter has been optimized, the inverted results are still greatly different from the real model. Finally, we suggest using the L1 norm, Kurtosis, variation, Cauchy norm with adaptive damping parameter or/and modified Cauchy norm with adaptive damping parameter as evaluation criteria to reduce the bad influence of inaccurate wavelet phase estimation and obtain good results in theory.

Phase estimation in bispectral domain based on conformal mapping and applications in seismic wavelet estimation

Seismic wavelet estimation is an important part of seismic data processing and interpretation, whose preciseness is directly related to the results of deconvolution and inversion. Wavelet estimation based on higher-order spectra is an important new method. However, the higher-order spectra often have phase wrapping problems, which lead to wavelet phase spectrum deviations and thereby affect mixed-phase wavelet estimation. To solve this problem, we propose a new phase spectral method based on conformal mapping in the bispectral domain. The method avoids the phase wrapping problems by narrowing the scope of the Fourier phase spectrum to eliminate the bispectral phase wrapping influence in the original phase spectral estimation. The method constitutes least-squares wavelet phase spectrum estimation based on conformal mapping which is applied to mixed-phase wavelet estimation with the least-squares wavelet amplitude spectrum estimation. Theoretical model and actual seismic data verify the validity of this method. We also extend the idea of conformal mapping in the bispectral wavelet phase spectrum estimation to trispectral wavelet phase spectrum estimation.

Incremental semi-supervised learning for intelligent seismic facies identification

Intelligent seismic facies identification based on deep learning can alleviate the time-consuming and labor-intensive problem of manual interpretation,which has been widely applied.Supervised learning can realize facies identification with high efficiency and accuracy;however,it depends on the usage of a large amount of well-labeled data.To solve this issue,we propose herein an incremental semi-supervised method for intelligent facies identification.Our method considers the continuity of the lateral variation of strata and uses cosine similarity to quantify the similarity of the seismic data feature domain.The maximum-diff erence sample in the neighborhood of the currently used training data is then found to reasonably expand the training sets.This process continuously increases the amount of training data and learns its distribution.We integrate old knowledge while absorbing new ones to realize incremental semi-supervised learning and achieve the purpose of evolving the network models.In this work,accuracy and confusion matrix are employed to jointly control the predicted results of the model from both overall and partial aspects.The obtained values are then applied to a three-dimensional(3D)real dataset and used to quantitatively evaluate the results.Using unlabeled data,our proposed method acquires more accurate and stable testing results compared to conventional supervised learning algorithms that only use well-labeled data.A considerable improvement for small-sample categories is also observed.Using less than 1%of the training data,the proposed method can achieve an average accuracy of over 95%on the 3D dataset.In contrast,the conventional supervised learning algorithm achieved only approximately 85%.

新工科建设视域下地球物理实践教学改革新探

在当前加强新工科建设的大背景下,我国对创新型卓越工程技术人才的培养选拔提出了新的要求,生产实践教育环节愈显重要。结合地球物理勘探专业生产实践教学特点与现状,针对影响实践教学质量的主要问题和因素,以培养广大学生的工程意识和提高学生的工程实践能力为目标,通过对策研究与初步实践,探索并提出了提高地球物理专业高素质人才培养质量的实践教学模式改革的新思路,建议高校在生产实践教学模式、实践教学内容与形式、指导教师团队、监督与评价制度等方面不断完善创新,形成面向应用的地球物理专业高质量工程实践教育模式。

Multi-well wavelet-synchronized inversion based on particle swarm optimization

Wavelet estimation is an important part of high-resolution seismic data processing.However,itis diffi cult to preserve the lateral continuity of geological structures and eff ectively recover weak geologicalbodies using conventional deterministic wavelet inversion methods,which are based on the joint inversionof wells with seismic data.In this study,starting from a single well,on the basis of the theory of single-welland multi-trace convolution,we propose a steady-state seismic wavelet extraction method for synchronizedinversion using spatial multi-well and multi-well-side seismic data.The proposed method uses a spatiallyvariable weighting function and wavelet invariant constraint conditions with particle swarm optimization toextract the optimal spatial seismic wavelet from multi-well and multi-well-side seismic data to improve thespatial adaptability of the extracted wavelet and inversion stability.The simulated data demonstrate that thewavelet extracted using the proposed method is very stable and accurate.Even at a low signal-to-noise ratio,the proposed method can extract satisfactory seismic wavelets that refl ect lateral changes in structures andweak eff ective geological bodies.The processing results for the fi eld data show that the deconvolution resultsimprove the vertical resolution and distinguish between weak oil and water thin layers and that the horizontaldistribution characteristics are consistent with the log response characteristics.