基于非下采样剪切波变换—参数自适应脉冲耦合神经网络的属性融合裂缝预测方法

Attribute fusion method based on NSST—PAPCNN for fracture prediction

常用的叠后地震属性主要有相干体(描述波形相似性)、曲率体(表征构造应力引起的地层弯曲程度)、倾角体(刻画地层构造变化特征)等,但仅仅依靠单一属性很难准确地预测地下裂缝分布情况。为此,提出一种基于非下采样剪切波变换(NSST)—参数自适应脉冲耦合神经网络(PA-PCNN)的属性融合裂缝预测方法,该方法基于NSST分解算法,将多种属性数据分解为高、低频子带,将融合后的多尺度、多方向高、低频子带进行数据重构,得到最终的多属性融合结果,可进一步提取裂缝的轮廓及细节信息。具体步骤为:①提取描述相同尺度裂缝的多种地震属性(相干、曲率及倾角等属性),通过NSST将多种属性分解为高、低频子带,其中高频子带包含更多的裂缝细节信息,低频子带可更好地刻画裂缝轮廓且具有丰富的能量信息。②对高频子带运用PA-PCNN模型进行融合,无需人工设置参数,得到更全面的高频数据;结合八邻域的改进拉普拉斯算子加权和与局部能量加权方法对低频子带进行融合,使低频数据更好地保留细节及能量信息,以得到丰富的低频数据。③通过逆NSST方法有效地完成属性融合裂缝预测。运用所提方法对M区属性数据进行测试,并对比了不同方法的属性融合裂缝预测结果,证明基于NSST—PAPCNN的属性融合裂缝预测方法能够更有效地预测裂缝。

The commonly used post-stack seismic attributes mainly include the coherent body(describing the similarity of waveform),curvature body(describing the bending degree of formation caused by tectonic stress),dip-angle body(describing the structural changes of formation).However,it is difficult to accurately predict the distribution of underground fractures only by a single attribute.Therefore,this paper proposes a comprehensive attribute fusion method based on nonsubsampled shear wave transform-parameter adaptive pulse coupled neural network(NSST-PAPCNN)for fracture prediction.This method relies on the NSST decomposition algorithm to decompose the multiple attribute data into high-and low-frequency sub-bands.After fusion,the multi-scale and multi-direction high-and low-frequency sub-bands were refactored.From the final multi-attribute fusion result,the contour and detail information of fractures can be further extracted.The detailed steps are as follows:①Multiple seismic attributes(coherence,curvature and dip-angle attributes)of the same-scale fractures were extracted and decomposed into high-and low-frequency sub-bands by NSST.The high-frequency sub-band contains more fracture information,and the low-frequency sub-band can better describe the fracture contour and has rich energy information.②PA-PCNN model was used for the fusion of high-frequency sub-band without manual parameter setting,which generated more comprehensive high-frequency data.The weighted sum of eight-neighbor modified Laplacian(WSEML)and the weighted local energy(WLE)were combined to fuse the low-frequency sub-bands,enriching the low-frequency data by retaining more details and energy information.③The inverse NSST method was applied to predict the fracture effectively based on attribute fusion.The proposed method was used to test the attribute data in M Zone,and the fracture prediction results of different methods are compared.It proves that the attribute fusion based on NSST-PAPCNN can predict fractures more effectively.