论文部分内容阅读
利用人工神经网络,可以由钻孔的孔隙度值预测同深度下的渗透率值。使用的是一种很适合进行预测的反向传播的网络结构。在传统的回归分析法预测渗透率的过程中,通常假定孔隙度和渗透率的关系是已知的。但实际上,这种函数形式即模式方程并不确定。而神经网络法并不事先对这种函数关系作任何假设。在亚拉巴马州南部的 Big Escambia Creek(BEC)侏罗纪 Smackover 碳酸盐岩油田,我们选择了6口井来检验该方法的应用效果。所用原始数据是各井的孔隙度及对应的空间坐标。预测过程可分3种情况进行,每种情况中,选择6口井的一个子集作为训练集,一口井为校准井,1～2口井用来预测。并用简单的线性回归预测结果与之比较。在第一种情况中,神经网络法的预测效果要明显好于回归分析;另外两种情况的预测效果等同。神经网络可利用较少的数据准确地预测渗透率,也易于参杂一些其它信息(如来自测井、岩心的岩类信息)。除此之外,碳酸盐岩储层的分割性也可用这种方法加以识别。
Using artificial neural networks, permeability values at the same depth can be predicted from borehole porosity values. Using a well-designed backpropagation network architecture. In the traditional regression analysis of permeability prediction process, it is generally assumed that the relationship between porosity and permeability is known. In fact, however, this functional form, the mode equation, is not certain. The neural network method does not make any assumptions about the relationship of this kind of function in advance. In the Big Escambia Creek (BEC) Jurassic Smackover carbonate field in southern Alabama, we selected six wells to test the application of the method. The raw data used is the porosity of each well and the corresponding spatial coordinates. The prediction process can be divided into three cases. In each case, a subset of 6 wells is selected as the training set. One well is a calibration well and 1 ~ 2 wells are used for prediction. And compared it with a simple linear regression prediction. In the first case, the prediction effect of neural network method is obviously better than the regression analysis; the other two cases have the same prediction effect. Neural networks can make use of less data to accurately predict permeability and also to miscellaneous some other information (such as rock information from logs and cores). In addition, the fragmentation of carbonate reservoirs can also be identified in this way.