基于PSO-K均值聚类的石脑油干点值多模型软测量建模方法

Multi-mode soft sensor modeling of naphtha dry point based on PSO-K means clustering

针对蒸馏工艺过程中变量多且变量间关系呈强耦合性及高度非线性的特点,提出一种基于PSO-K均值聚类的石脑油干点值多模型软测量建模方法。分析石脑油干点值的影响因素,采集相关辅助变量。用基于PSO改进的K均值聚类算法将现场采集数据进行划分,得到样本子集。再将得到的各个样本子集分别用SVM算法进行训练,建立石脑油干点值的预测子模型。在此模型基础上,通过计算预测样本与各子模型训练样本聚类中心的欧氏距离,采用模型切换的方法选择预测模型。仿真结果表明该方法避免了分类时K均值算法易陷入局部极值的问题,可以有效预测常压塔石脑油干点值,与单个全局模型相比有更好的精度与泛化能力。

A multi-mode soft sensor modeling of naphtha dry point based on PSO-K means clustering is developed according to the strong coupling and highly non-linear characteristics between the multi variables during the distillation process. The factors that influence naphtha dry point are analyzed, and relevant auxiliary variables are collected. Subset samples are gained through partitioning the field data with K-means algorithm clustering which improved by PSO first. Then the predictive sub-models are established by training samples subset with SVM algorithm respectively. The Euclidean distances between predictive samples and the training samples of the each sub-model are computed based on the models, and select predictive models by switching the models. The simulations reveal that the method can avoid the K means algorithm droping into local optimum easily, and is effective enough to predict the naphtha dry point of the atmospheric tower. In comparison with the single global model this method has higher precision of prediction and better capacity of generalization.