地震倒谱特征参数谱聚类地震相分析方法

Seismic facies analysis based on cepstrum characteristic parameters and spectral clustering

常用的地震相分析方法有随机模拟、神经网络、聚类算法和深度学习等。随机模拟结果易受随机模型影响,而且在地质结构复杂地区难以准确划分地震相。神经网络和深度学习具有较强的容错性和泛化能力,但需要海量训练样本数据,同时训练网络的计算量巨大。K均值聚类、C模糊聚类等经典聚类算法在简单数据集上均获得了理想的聚类结果,但对于非凸数据集并不能实现全局最优。为此,提出一种基于地震数据倒谱特征参数的谱聚类地震相分析方法。该方法以地震倒谱特征参数为谱聚类的输入变量,然后通过井标定,建立地震相与地质体间的对应关系。以图论为基础的谱聚类方法将数据的聚类转化为图的分割问题,通过图的最优分割实现数据的精确聚类。通过优化相似度矩阵计算方法,构建稀疏相似度矩阵,可以解决矩阵维度过大引起的存储和计算量大的问题,使谱聚类更适用于划分三维空间地震相。地震倒谱特征参数一方面能减少数据的维数,降低计算复杂度;另一方面能消除波形的影响,提高划分精度。模型试验和实际数据应用表明,与地震瞬时振幅、多地震属性地震相划分结果相比,所提方法划分的地震相带与古地貌吻合更好,边界更清晰,可解释性也更好,可为油气勘探和油藏评价提供数据支撑。

Stochastic simulation,neural network,clustering and deep learning are always used to seismic facies analysis.However,stochastic simulation results are easily affected by stochastic models,and it is difficult to get accurate seismic facies division in complex geological areas.Neural network and deep learning methods have strong fault tolerance and generalization ability,but they require massive training samples and high computing costs.Classical clustering algorithms such as K-means clustering and C-fuzzy clustering can obtain ideal clustering results on simple data,but cannot achieve global optimization for non-convex data.To overcome these problems,we propose a seismic facies analysis method based on cepstrum characteristic parameters and spectral clustering.In this method,seismic cepstrum characteristic parameters are calculated as input variable of spectral clustering,and then the corresponding relationship is established between seismic facies and geological body after calibrating by well data.The spectral clustering method based on graph theory transforms data clustering into graph segmentation,which achieve accurate clustering through optimal graph segmentation.And we also construct a sparse similarity matrix through optimizing the similarity matrix calculation method,by which the storage and calculation problems caused by large matrix dimensions can be solved.Therefore,spectral clustering is more suitable for 3D seismic facies division.The advantages of cepstrum characteristic parameters are as follows:on the one hand,it can reduce data dimension and computational complexity;and on the other hand,it can eliminate the influence of waveform,and improve division accuracy.The applications to model and real data show that the seismic facies divided by the proposed method are in better agreement with the paleogeomorphology than the facies division based on instantaneous seismic amplitude and multiple seismic attributes,showing clearer boundaries and better interpretability.The results are reliable data for oil exploration and reservoir evaluation.