BP神经网络隐式法在测井数据处理中的应用  

Processing of well log data based on backpropagation neural network implicit approximation

作　　者：李道伦[1] 卢德唐[2] 孔祥言[2] 杜奕[2] 

机构地区：[1]中国科学技术大学计算机科学与技术系,安徽合肥230026 [2]中国科学技术大学安徽省计算与通讯重点实验室

出　　处：《石油学报》2007年第3期105-108,共4页Acta Petrolei Sinica

基　　金：国家重点基础研究发展规划(973)项目(2006CB705800);西南石油大学油气藏地质开发工程国家重点实验室开放基金项目(PLN0409)资助

摘　　要：现有神经网络方法对时间向量序列数据的处理是通过单点进行的,割裂了数据间的关联性。为此,利用隐式曲线的构造原理,通过对时间向量序列的变换,提出了一种整体预测时间向量序列的测井数据的方法。神经网络隐式整体预测方法的步骤是:①将数据变换为封闭曲线,构造约束点以简化神经网络的输入与输出;②利用神经网络的隐式方法,通过智能学习和仿真模拟,得到封闭的预测曲线;③经过变换得到最终的预测曲线。实验证明了该方法的有效性。The previous methods based on neural network are difficult to predict the water saturation of next year based on the data of the given years. A new method combining the neural networks with the principle of implicit curve can effectively handle the above problem. First, the vector data of every year are mapped into a closed curve, and a virtual explicit function is constructed on the constraint points. Then, the explicit function is approximated by a backpropagation neural network. Finally, the isoline of the neural network is extracted from the simulation surface. The predicted data can be obtained by the inverse mapping of the isoline. Some experiment results verified the effectiveness of this method.

关 键 词：BP神经网络 隐式曲线 测井数据 预测方法 时间向量序列 测井曲线 数值模拟 

分 类 号：TE319[石油与天然气工程—油气田开发工程]