论文详情
火成岩储层岩性复杂，识别难度大，当已知地层信息较少时，传统的交会图和有监督神经网络（如BP神经网络）等方法在识别岩性时会受到一定限制。为此，基于自组织神经网络（SOM网络)的结构和原理，在松辽盆地南部利用实际测井资料建立了火成岩样本数据集；利用SOM网络对样本数据集进行了训练，得到了数据集的聚类结果；讨论了SOM网络的标准化方式、结构参数和测井曲线对聚类结果的影响，认为利用正态标准化方法、选择合适的结构参数和测井曲线，以样本数据集的聚类结果作为分类基础，对火成岩井段测井资料进行了岩性识别，获得了较好的效果。
The lithology is rather complex and difficult to identify in igneous reservoirs. With little formation information, traditional cross-plot and supervised neural networks (such as BP network) are restricted in identifying lithology. Therefore, in the south part of SongLiao basin, based on the principles and structure of SOM neural network, the data set of igneous samples were established by actual logging data. The cluster results were obtained by training the samples with SOM network. the influence of standard means, structure parameters and log of SOM network on cluster results, which shows that good results can be achieved for lithology recognition on logging data of igneous reservoir by using normal standard method, selecting proper structure parameters and log, and taking the cluster results as the basis of classification.