采用支持向量回归从测井曲线定量计算油页岩含油率

QUANTIFYING OIL CONTENT OF OIL SHALE FROM WELL LOGS USING SUPPORT VECTOR REGRESSION

获取油页岩含油率是油页岩资源评价的重要步骤,传统从测井曲线计算油页岩含油率多采用回归模型,但存在误差大或过拟合的局限性和弱点.本文尝试结合大数据概念的数据挖掘算法和测井应用知识进行油页岩含油率定量计算,提高含油率计算的精度以及模型的泛化性.利用改进的ΔlogR技术获得DT_s、DEN_s、GR_s作为解释变量.采用数据挖掘算法——支持向量回归进行定量计算油页岩含油率能够大幅提高泛化性和精度,获得模型训练样本R^2得分为0.82,测试样本R^2得分可达0.70,拟合精度较高.支持向量回归模型比传统回归模型泛化能力更强,能够避免过拟合问题,具有广泛的应用性.

The oil content is an important evaluation index for oil shale resources. Traditionally, calculation of the oil content from well logs of oil shale is performed by regression model, which, however, has the limitation and weakness of big error or over-fitting. This paper attempts to integrate the classical data mining algorithm with ＂Big Data＂ concept and logging application knowledge for oil content quantification to improve the accuracy and generalize the model. The explanatory variables of DTs, DENs and GRs for analysis are obtained by the improved AlogR technique. The data mining algorithm of support vector regression （SVR） can greatly improve the model generalization and precision in the oil content quantification. The R2 score of training samples in the model is 0.82. A high fitting precision is achieved in the test samples, of which the R2 score is 0.70. The SVR model and can avoid over-fitting problem and well applied. is more generalized than traditional regression model,