基于偏最小二乘回归法的储层厚度预测

Reservoir thickness prediction based on partial least-squares regression method

首次采用偏最小二乘回归法进行储层厚度预测,推导其数学算法,总结其优势,建立正演模型分析其可行性。针对靶区的地震数据进行地震属性的提取,优选出可以较好地描述砂体分布情况的5种属性,分别为波峰数、平均振幅、平均瞬时相位、振幅立方差和能量半衰时。利用这5种属性分别对靶区应用主成分分析法、神经网络法和偏最小二乘回归法,得到井点处的砂体厚度预测值。根据各自绝对误差和相对误差,推断应用最小二乘回归法预测砂体厚度值更为准确。根据建立的回归方程,对靶区进行砂体厚度预测,得到砂体厚度分布情况。

On the basis of deducing the mathematical algorithm and summarizing the advantages,the feasibility of the partial least square regression method,to be the first in predicting reservoir thickness,was analyzed by building up the forward model.According to the seismic data of target areas,five kinds of seismic attributes,which can describe the sand body distribution well,were extracted. The five kinds of seismic attributes include the number of peaks,mean amplitude,average instantaneous phase,amplitude variance and energy half- life. Based on the five kinds of seismic attributes,the sandbody thicknesses of some wells in the study area were respectively predicted by adopting the principal component analysis method,the neural network method,and the partial least squares regression method. It was found in the comparison of the corresponding absolute and relative errors that the value predicted by the least squares regression method was more accurate. Based on the established regression equation,the sandbody thickness in the target area can be forecasted to obtain the sandbody thickness distribution.