论文部分内容阅读
目前，利用神经网络划分地震微相的方法可以归纳为两大类：一类是监督型模式识别；另一类是非监督型模式识别。当样本资料不足或不具代表性时，采用监督型模式识别容易造成强行分类而导致错误结论。以Kohonen网络为代表的非监督型模式识别的缺陷是：①不能动态聚类；②输入模式动态范围较大时，量值小的模式不易识别；③最终炎别关系亲近程度不清。本文对非监督型模式识别作了改进，应用改进后的识别方法，可以实现由细到粗的类别动态系类，而且最终类别间亲近关系也非常明显。
At present, the method of dividing seismic micro-facies by neural network can be summarized into two categories: one is supervised pattern recognition and the other is unsupervised pattern recognition. When the sample data is not enough or not representative, using supervised pattern recognition easily leads to forced classification and leads to wrong conclusions. The disadvantages of unsupervised pattern recognition represented by Kohonen network are as follows: ①Can not be dynamically clustered; ②When the dynamic range of input mode is large, the pattern of small value is not easy to identify; ③The relationship between the final inflammatory relationship is unclear. In this paper, the unsupervised pattern recognition is improved. By using the improved recognition method, it is possible to realize the classification of fine to coarse categories, and the close relationship between the final categories is also very obvious.