基于核空间的模糊聚类方法在储层预测中的应用

Application of kernel fuzzy C-means method to reservoir prediction

基于核空间的模糊C均值聚类方法是一种模式识别的新方法。在地震属性聚类处理时常常会遇到非超球体数据以及非线性类间边界等问题,而传统的模糊C均值聚类方法无法行之有效地解决。将核空间方法引入传统的模糊C均值聚类方法中,并应用于储层预测。针对地震属性聚类问题中不同属性对于储层的敏感性不同,将特征权重和模糊指数等参数加以优化,提高新的模糊聚类方法的储层预测效果。对实际资料的计算与分析结果表明,新的基于核空间的模糊C均值聚类方法可以更准确地刻画碳酸盐岩含气储层边界。

The kernel fuzzy C-means（FCM） method is a novel method for pattern recognition.The problems such as non-hyperspherical data and non-linear inter-class boundary are prevalent during seismic attributes clustering process,which could not be resolved effectively by traditional FCM method.The kernel function was introduced into traditional FCM method for these problems in reservoir prediction.The parameters including feature weights and fuzzy coefficient were optimized for different sensibility of seismic attributes,which could improve the effectiveness of this new kernel FCM method for reservoir prediction.The results of experiments on the artificial and real data show that the new kernel FCM method can describe the boundaries of gas-bearing carbonate reservoir more accurately.