基于模式识别和谱图映射的通用油品调合模型

传统油品调合优化模型建立质量约束时依次建立各个性质的卡边约束，具有非线性调合关系的性质无法用统一的调合模型来描述，工艺条件变动导致组分油波动时调合模型就会产生较大的偏差。针对该问题，提出了一种基于油品近红外谱图的通用建立质量约束的方法，将非线性不统一的性质约束转化为特征空间上得分向量间的线性约束。根据组分油在特征空间的投影用朴素贝叶斯分类器判别到训练集的某类样本中，用核密度估计方法在成品油类中确定一个约束区域，然后利用组分油和成品油得分向量之间的线性加和关系建立调合油品的质量约束。这种建模方法能适应工艺条件的波动，适合不同的炼油厂，也适用于汽油、柴油、原油等不同油品的调合。用某炼油厂的柴油数据进行调合实验验证了该方法的有效性和准确性。

General Oil Blending Model Based on Pattern Recognition and Spectra Projection

 The control objectives for traditional oil blending optimization models are framed in terms of meeting specifications on properties of the product oil. However, there is no universal blending model suitable for illustrating the property blend in a nonlinear fashion. Along with the varying of blend components, the blending model becomes inaccurate. To solve this problem, we presents a novel method based on the Near infrared (NIR) spectra of oil, which converts nonlinear property constraints to linear constraints of scores in feature space. Project each blend component into the feature space and use Naive Bayes classifier to classify it into one group for the training samples. The kernel density estimation method is used to formulate a constraint region in the product oil samples. Finally, the quality constraints based on the linear relationship between the product oil score and components scores are formulated. This method is widely suitable for different refineries and different oil blending. The experiment of a diesel optimization problem validates the efficiency and accuracy of this method.