基于局部熵的LP-PCNN多属性融合裂缝预测方法

LP-PCNN multi-attribute fusion fracture prediction method based on local entropy

地震属性分析方法在储层裂缝预测方面是常见且有效的方法,但地震属性与裂缝之间往往是多元复杂的非线性关系,单属性分析的结果存在不稳定和多解性问题。为了更加全面、准确地预测储层裂缝特征,提出了一种基于拉普拉斯金字塔算法(LP)和脉冲耦合神经网络(PCNN)的多属性融合分析方法。通过属性分析提取并优选对裂缝敏感的多个单属性,引入拉普拉斯金字塔算法,在保护高频细节信息的前提下将各单属性分解为多尺度空间频带;利用PCNN模型强大的非线性处理功能对分解数据进行聚类特征分析;引入表征统计特性的局部熵(LE)对拉普拉斯金字塔的各个分解尺度进行融合。利用拉普拉斯金字塔重构算法得到最终多属性融合结果。实际地震资料测试结果表明,该方法不仅能够有效整合裂缝信息,更加全面地预测裂缝展布特征,而且能在一定程度上压制单属性中存在的冗余或错误信息,提高信噪比,对裂缝边界的刻画更加清晰。

Fractures improve the reservoir space performance of oil and gas,and provide important channels for oil and gas migration.The degree of development and distribution of fractures affect the production volume and stability of tight sandstone reservoirs.Seismic attribute analysis is a common and effective method in reservoir fracture prediction;however,the relationship between seismic attributes and fractures is often multivariate,complex,and nonlinear.The results of single attribute analysis may be unstable and have multiple solutions,which make it difficult to predict reservoir fractures accurately.To predict reservoir fracture characteristics more comprehensively and accurately,a multi-attribute fusion method based on the Laplace pyramid(LP)algorithm and pulse-coupled neural network(PCNN)is proposed.Multiple single attributes which are sensitive to fractures were obtained based on seismic attribute analysis.To protect high frequency details,the LP algorithm decomposes each individual attribute into multi-scale spatial frequency bands.The powerful nonlinear processing function of the PCNN model is used to analyze the clustering characteristics of the decomposition data.When directly using the ignition frequency of each sampling point,attribute fusion shows one-sidedness and high sensitivity to edges.To avoid this,local entropy(LE),which represents statistical characteristics,is introduced to fuse each LP decomposition scales.The PCNN model has a powerful nonlinear processing function which couples the influence of surrounding neurons and suppresses redundant information within a single attribute.The final multi-attribute fusion result is obtained by reconstruction.The experimental results show that the proposed method can improve the signal-to-noise ratio,predict the fracture distribution characteristics more comprehensively and effectively,and delineate fracture boundaries more clearly.