岩屑荧光图像的混合石油组分研究

Mixed oil components analysis from cutting fluorescent image

石油是一种以烃类和非烃类化合物为主的复杂混合物,其主要的三种组分为:油质组分、胶质组分和沥青质组分.通过采集混合石油岩屑荧光图像中不受采集参数(相机曝光时间与光圈值)影响的归一化颜色特征分量rgb以及归一化色度分量h,利用最小二乘回归、支持向量回归、局部加权投影回归对这些特征和已知混合石油各个组分的含量进行训练,最后利用这些回归模型对未知组分的混合石油进行预测.经测试,局部加权投影回归的效果最好,对于油质组分,其预测结果的均方根误差为13.50%;对于胶质组分,其预测结果的均方根误差为13.34%;对于沥青质组分,其预测结果的均方根误差为7.03%.实验表明,预测结果较好,能够满足实际应用的需求.

Oil is a kind of complex mixture which consists of hydrocarbons and nonhydrocarbons. It is composed of three main components： oiliness, resin and bitumen. After acquiring the normalized rgb value and the normalized hue value which are insensitive to the acquiring parameters（the exposuretime and the aperture of the camera） from the oil cuttings fluorescent image, Least Square Regression（LSR）, Support Vector Regression（SVR）, Locally Weighted Projection Regression（LWPR） are used to train these features and the known percentage of each components in the mixed oil, at last the trained regression models are made use of to predict the unknown percentage of other components. The results show that the LWPR is the best regression forecasting model,and its RMSE of oiliness regression forecast is 13.5%, its RMSE of resin regression forecast is 13.34% and its RMSE of bitumen regression forecast is 7.03%. It shows that the predicted results are good enough to meet the needs of practical application.