基于小波神经网络的ODP1148A井岩性预测  

The lithology prediction of ODP hole 1148A based on the wavelet neural network

作　　者：王辉[1] 黎明碧[2] 唐勇[2] 方银霞[2] 丁巍伟[2] 付洁[2] 王聪[3] 

机构地区：[1]中国石油集团测井有限公司长庆事业部,西安710201 [2]国家海洋局海底科学重点实验室,杭州310012 [3]浙江大学地球科学系,杭州310012

出　　处：《地球物理学进展》2014年第1期392-399,共8页Progress in Geophysics

基　　金：国家基础研究发展计划项目("973"项目)子课题"南海大陆边缘新生代层序地层与沉积演化"(2007CB41170301);国家海洋局海洋公益性行业科研专项资助项目(201205003;200805078;JT1306)联合资助

摘　　要：为对大洋钻探计划(ODP)1148A井岩性收获率低的层位和涂片分析之外的井段进行岩性预测,得到更全面的岩性信息,设计出了一个以"morlet"小波为隐含层传递函数的三层小波神经网络.将总样本中的一部分作为学习样本,用于小波神经网络的训练,另一部分作为测试样本,用于检验小波神经网络预测岩性正确与否以及预测结果的误差评价.经过对该小波神经网络反复测试和调整,最终得到了一个误差最小的神经网络,其测试结果显示岩性预测符合率60%以上的占总体的60%以上.在没有岩心资料或取芯收获率低的层段,可将其用于粗略的岩性参考.将经测试后较满意的小波神经网络应用于整口井的岩性预测,得到该井的详细岩性信息.该岩性预测结果弥补了岩心涂片分析数椐较少,以及该井岩心收获率低的层位的岩性空白.本文将小波神经网络应用于测井岩性预测方法的探讨可为人工智能与地球物理测井相结合提供新的思路.In order to predict the lithological characters of the low core recovery intervals of the hole 1148A of the Ocean Drilling Program （ODP） and of the intervals beyond the smear analysis, and to obtain more detailed lithology information, a wavelet neural network of three layers with ＇morlet wavelet as the transfer function of the hidden layer was devised and programmed based on the MATLAB software. The lithological smear samples of the ODP site 1148A were divided into two parts. One part of the total samples was designed as the learning sample used to train the net, and the other part was employed as the test sample planed to test whether the predicted lithology was true and to evaluate its error. After testing and adjusting the designed wavelet neural network over and over again, finally a most appropriate one was proposed with the right parameters and the minimum error. The test results showed that those with the lithology prediction coincidence rate of 60~ and above accounted for more than sixty percent of all the test samples. These results can be used as a rough lithology reference at intervals with no core or low core recovery rate. When this optimal wavelet neural network was applied to the lithology prediction for the ODP hole 1148A, the lithological information of the whole hole was obtained, including sand, silt and clay contents at all intervals in this hole. This result was a supplementary to the relatively rare smear analysis data and it filled in the blanks of the low core recovery intervals of this hole. The method of applying the wavelet neural network to the log data for lithology prediction can provide a new idea for the combination of the artificial intelligence and the geophysical well logging.

关 键 词：大洋钻探计划(ODP) 测井 小波神经网络 岩性预 测 南海北部 

分 类 号：P631[天文地球—地质矿产勘探] P738[天文地球—地质学]