针对成像测井资料上的裂缝、溶蚀孔洞和燧石结核地质现象在成像测井资料上的形态和分布差异,提出一种用梯度方向直方图统计量(HOG)和熵相结合计算成像测井资料上3种地质现象的特征量、并采用非线性的BP神经网络对这3种地质现象自动识别的方法.首先计算成像测井资料上3种地质现象梯度方向统计量的U1～U6和熵U7,并将其作为特征量,然后分析特征量对这些地质现象的区分性和敏感性,最后采用BP神经网络方法对塔河油田11口井的435个样本分为学习样本和测试样本进行学习回判和测试识别.试验结果表明,对裂缝有区分性和敏感的特征量有5个,分别为特征量U1、U3、U4、U5和U6;对溶蚀孔洞有区分性和敏感的特征量是U2;对燧石结核有区分性和敏感的特征量是U7.BP神经网络对221个学习样本中裂缝、溶蚀孔洞及燧石的回判正确率均为100％;对214个测试样本,BP神经网络燧石结核识别正确率为85.5％,裂缝识别正确率为88.5％,溶蚀孔洞识别正确率为84.0％.%For the differences of morphology and distribution of three kinds of geological phenomena in imaging logging data,including fractures,pores,and chert,a automatic recognition was proposed in the paper,which can calculate the characteristic parameters of these geological phenomena combined with gradient direction histogram statistics(HOG) and entropy,and can automatically recognize these three geological phenomena by using nonlinear BP neural network.First,calculate the three kinds of statistic gradient direction of geological phenomena of imaging logging data on the U1 ～ U6 and U7 entropy as the characteristic parameter,then,analyse the distinguish capacity and sensitivity of characteristic parameters on geological phenomena,finally,using BP neural network method,435 samples of 11 wells in Tahe oilfield were divided into learning and testing samples,to make back judgement and identification.The results of test show that,there exist five characteristic parameters which are sensitive on cracks,respectively U1,U3,U4,U5 and U6;U2 is sensitive on pores,and U7 is sensitive on flint.Taking 221 learning samples as researching objects,the discriminant accuracy rate of identifying fractures and pores and flint was 100％ by using BP neural network;For 214 testing samples,the correct rate of recognition of flint by using BP neural network is 85.5％,the correct rate of crack identification is 88.5％,the correct rate of pores recognition is 84.0％.
展开▼
