神经网络在油气层横向预测和地震道编辑中的应用

Application of neural network to reservoir lateral prediction and trace editing

1. 江汉石油学院, 江陵434102;2. 华北油田物探公司

Dept.of Gcophysics, Jianghan Pctrolcum Institutc, Jiangling City, Internal Postcodc 434102

本文绘出了用多层感知器(MLP)作油气层横向预测及地震道自动编辑的方法。作油气层横向预测时, 只需给网络提供叠偏地震剖面数据和标定井油气信息, 对原始数据无其它要求。经塔里木盆地和中原油田实际地震资料检验表明, 网络预测结果与钻井等资料相吻合。作自动编辑时, 由于用监督学习型网络MLP作分类器, 使系统具有良好的适应性, 并能充分反映用户的意愿。华北实际地震资料的计算结果表明, 编辑精度达96．5％。

Using multilayer perceptron (MLP) for rcservoir lateral prediction and seismic trace editing is presented in this paper. All one has to do in lateral prediction for reservoir is to input stacked migrated seismic data and calibrated well data into a network. The tests on real data ftom the Tarim Basin and Zhongyuan oil fleld exhibit that the predictions resulting from the neural network are in agreement with thc drilling data. As for trace editing,since a supervised learning network MLP is used as a classifier, the systcm is then provided with good applicability to meet user's needs. Calculations using data from Huabei Area show that the editing precision is as high as 96.5%.