【摘要】 为了提高油页岩含油率近红外光谱分析建模的预测精度和稳定性,开展了基于最小二乘支持向量机(LS-SVM)建模方法的对比研究.采用主成分-马氏距离(PCA-MD)和基于蒙特卡洛采样(MCS)2种方法进行了奇异样本的检测,采用径向基核函数的LS-SVM、偏最小二乘(PLS)和反向传播神经网络(BPANN)3种方法进行建模方法对比.结果表明,对于64个油页岩岩芯样本,与PCA-MD方法相比,采用MCS方法剔除奇异样本后所建PLS模型的预测精度提高了28%.对于MCS方法剔除奇异样本后的58个样品,采用KennardStone法划分了44个样品的校正集和14个样品的预测集,采用2阶导数和标准化预处理方法,建立了100个LS-SVM的校正模型,模型的预测决定系数R2平均值达到0.90以上,高于PLS和BPANN模型的对应值;且R2的变化量(0.02)小于BPANN模型的对应值(0.32).因此,MCS奇异样本检测结合LS-SVM方法可提高油页岩含油率样本建模的精度和稳定性.更多还原

【Abstract】 In order to improve the prediction accuracy and precision of near-infrared( NIR) spectroscopy model for analyzing the oil yield from oil shale,sixty-four oil shale samples from the No. 2 well drilling of Fuyu oil shale base were analyzed based on least squares support vector machines( LS-SVM) calibration models.The Principal component-mahalanobis distance( PCA-MD) method and the Monte-Carlo sampling-based detection of outliers( MCS) method were investigated as means of removing the outliers. The modeling methods of radial basis function-based LS-SVM, partial least squares( PLS) and back propagation neural network( BPANN) were compared. The results showed that,compared with PCA-MD,the prediction accuracy of PLS models based on MCS was improved by 28%. The samples after eliminating the outliers were divided into the calibration set with 44 samples and the prediction set with 14 samples using the Kennard Stone method. One hundred LS-SVM calibration models were established based on preprocessing method of second-derivative and autoscaling. The mean determination coefficient( R2) were more than 90% and higher than PLS and BPANN models,and the fluctuation of R~2 were less than BPANN models. Thus,LS-SVM regression with MCS method can improve the accuracy and precision of oil yield of oil shale modeling.更多还原