利用内插神经网络从测井资料中估算渗透率

• 作 者: P. M. Wong 何婷婷 金燕
• 机 构: 四川石油管理局地质勘探开发研究院
• 出 处: 天然气勘探与开发
• 发表日期: 1999年02期
• 关键词: 渗透率 神经网络 岩石物理学 BPNN 内插 梯度下降法 天然气勘探 非均质 训练样本 岩相

对岩石物理学家而言,利用测井资料对储层渗透率进行估算是项艰难的工作。许多研究已经表明:由于逆向传播神经网络(BPNN)具有学习和推广能力,因而它是迄今为止估算渗透率的一种最有前途的方法。而且在逆向传播学习算法中利用梯度下降法常常可提供均匀估算值。本文提出一种固有的非均匀解,即内插神经网络(INET)。并通过一实例研究对比了两种神经网络的性能,其结果表明:对于检测井,INET提供比BPNN方法更精确、固有的非均匀渗透率剖面。同时该方法在计算机上也易于实现。

For rock physicists, using logging data to estimate reservoir permeability is a difficult task. Numerous studies have shown that BPNN is one of the most promising methods for estimating permeability so far due to its learning and promotion capabilities. Moreover, the use of gradient descent methods in backpropagation learning algorithms often provides uniform estimates. In this paper, an inherent inhomogeneous solution is proposed, namely the Interpolation Neural Network (INET). An example is given to compare the performance of the two neural networks. The results show that INET provides a more accurate and inherently non-uniform permeability profile than the BPNN method for the detection wells. At the same time the method is also easy to implement on the computer.