随机森林回归在地震储层预测中的应用

针对储层预测的复杂非线性及稳定性问题，将随机森林回归算法引入到地震储层预测中，建立地震属性与储层特征参数之间的非线性关系。以多种不同的地震属性为基础，通过构建井旁道地震属性与特征参数的回归森林模型进行储层预测，以预测值与实际值之间的均方根误差值为评价标准，分析随机森林回归算法在地震储层预测中的特点。将本方法应用于某陆地工区的自然电位预测和某海上工区的自然伽马预测，并与支持向量回归机方法的预测结果进对比，结果表明，尽管地震数据受到较强噪声的影响，随机森林方法仍可较好地刻画出储层的三角洲前缘沉积特征，表现出较好的稳定性和较高准确性。

Application of random forests for regression to seismic reservoir prediction

Aiming at the nonlinear and stability issue of reservoir prediction,we introduced random forests for regression algorithm to seismic reservoir prediction by constructing the nonlinear relationship between seismic attributes and reservoir feature parameter.Several different kinds of attributes used as foundation,we predict reservoir parameter through fitting seismic attributes data along wells with feature parameter,regarding the root-mean-square error of predicted values and real values as evaluation standard,and then analyze the character of random forests regress algorithm applied in seismic reservoir prediction.This method is applied to predicting spontaneous potential for an inland survey and predicting natural gamma for an offshore survey.Meanwhile,we compare the prediction results with supported vector regression machine.The comparing result reveals that although the seismic data affected by strong noise,random forests method had better depiction of delta front depositional feature and perform better stability and accuracy.