基于交叉验证的地震多属性概率神经网络(PNN)反演在识别热瓦普地区火成岩中的应用

Multi-attribute probabilistic neural network inversion applicated in identifying igneous in RWP area based on cross-validation

新疆塔北地区发育巨厚二叠系火成岩,速度差异较大,而且火山喷发模式难确定,给其下伏低幅度碎屑岩圈闭和岩性圈闭落实带来困难。本文对二叠系火成岩利用概率神经网络反演等方法进行精细的速度场研究。概率神经网络反演是一种典型的非线性反演方法,相比于稀疏脉冲反演,在地震反演过程的非线性问题,具有更好的分辨率。通过逐步回归和交叉验证优选使验证误差最小的属性组合,使反演结果与测井属性有更好的相关性。建立的速度场经验证,更符合火成岩分布与速度变化规律。

There developed huge thick Permian Igneous in Tabei area of Xinjiang.The Permian igneous rocks with sharp variation of velocity,affects the process of oil and gas exploration seriously,and makes trap-confirming more difficult.For solving this problem,this paper use Probabilistic Neural Network inversion method to establish igneous velocity field.Compared with CSSI,PNN inversion is a typical nonlinear inversion with its high resolution.At first,agroup of attributes was selected by using Stepwise regression and cross-validation for analyzing and error minimum,to make inversion results have better correlation with log properties.The inversion velocity field was testified to conform the distribution of igneous and velocity changes.