Abstract: Assuming that each seismic trace is taken as a discretized time series and a multi-dimensional sample space with various informations, autoregression technique is used, under a condition of minimum loss of information, to find out the space characteristic factor of each sample and thus the number of dimensions is largely decreased, characteristic factors of all seismic traces in a layer are estimated,and are classified by means of multiple distance principle. On this basis, fuzzy cluster technique is used to clusterize all the parts classified in a layer, then, in-layer lateral lithological variation can be divided and classified.Theoretical model is used to verify this technique and this technique is used to process an actual seismic section, and satisfactory results are obtained.

摘要： 假设地震记录道可视为一个离散的时间序列,并认为它是一个包含各种信息的高维样本空间,利用自回归模式识别方法,在信息量损失最小的条件下,求取样本空间的特征系数,可使维数大大降低,按层计算出所有地震道的特征系数,利用多种距离准则,在层内对它们进行划分.对划分出的各部分利用模糊聚类方法进行聚类,实现了层内岩性横向变化的划分和分类.用理论模型对本方法进行了验证,并将它应用于实际地震剖面处理,获得了满意的效果.

关键词: 自回归, 模式识别, 特征系数, 模糊聚类, 距离准则, 岩性变化, 地震剖面