基于核函数的非线性校正算法在原油快评中的应用

Application of Nonlinear Algorithm Based on Kernel Methods in Fast-Evaluation of Crude Oil

为了快速准确地测量原油的密度、酸值和硫质量分数等重要性质,采用红外光谱技术结合非线性化学计量学定量校正算法建立校正模型。结果表明,分别使用最小二乘支持向量机算法(LSSVM)和核偏最小二乘(KPLS)两种基于核函数的非线性校正算法建模预测原油密度、酸值和硫质量分数的预测标准偏差分别为0.0065g/cm3、0.19mgKOH/g和0.38%以及0.0089g/cm3、0.23mgKOH/g和0.40%,预测结果的重复性与再现性等同或优于标准方法。与经典偏最小二乘(PLS)方法相比,KPLS算法准确性更高,而LSSVM具有更快的训练速率、更小的测量偏差等优点。

Fast and accurate measurement of density, acid value and sulfur mass fraction is necessary for petroleum characterization, which can be realized by the combined use of mid-infrared spectroscopy and nonlinear quantitative calibration algorithm based on kernel function. The calibration models of density, acid value and sulfur mass fraction were established by Least squares support vector machines （LSSVM） and Kernel partial least squares （KPLS） with the standard prediction errors （SEP） of crude oil density,and sulfur mass fraction were 0.0065 g/cm^3 ,0. 19 mgKOH/g and 0. 38% on LSSVM and 0. 0089 g/cm^3,0. 23 mgKOH/g and 0. 39% on K P L S, respectively. The results predicted by the two methods were very close to those determined by standard methods. Compared with classic PLS algorithm, the KPLS showed high predictive accuracy, and LSSVM method provided the advantages such as high-speed, simplicity and high precision.