直馏减压馏分运动黏度的预测

PREDICTION FOR KINEMATIC VISCOSITY OF VACUUM GAS OIL

采用逐步线性回归和人工神经网络(ANN)方法,建立了基于烃类组成的40℃和100℃的直馏减压馏分(VGO)运动黏度的预测公式;讨论了不同回归因子和不同回归方法对模型建立的影响。逐步回归方法可以给出直观的数学公式,并且能够了解各烃类对VGO运动黏度的影响。人工神经网络方法对非线性关系的物理量具有较强的预测能力,可以根据不同的需要选择不同的方法建立预测模型。采用成对t检验方法对VGO运动黏度模型进行了检验,并分析了造成偏差的原因。

The prediction model for the kinematic viscosity of vacuum gas oil at 40℃ and 100℃ was established based on the detail hydrocarbon types by stepwise regression method and artificial neural network （ANN）. The effects of the factors, such as regression factor and regression method, on the prediction model were discussed. The stepwise regression method was simple in form, and could directly reflect the hydrocarbon type influence on the kinematic viscosity of VGO. The ANN models had good ability to predict the variables with nonlinear trend, and different ANN models could be selected to get the prediction model according to the different needs. The prediction model was examined by paired t test, and the reason of bias was discussed.