PLS模糊多模型建模方法在航煤干点估计中的应用研究  

The application study of fuzzy multi-model modeling method based on partial least squares for estimation of the dry point of aviation kerosene oil

作　　者：孟凡强[1,2] 于佐军 

机构地区：[1]中国石油大学(华东)信息科学与工程学院 [2]济南军区青岛警备区71271部队

出　　处：《计算机与应用化学》2012年第7期885-889,共5页Computers and Applied Chemistry

摘　　要：针对常压塔复杂工况下的航煤干点估计困难的问题,本文提出一种基于PLS模糊多模型软测量建模方法:(FuzzyMulti-model based on PLS,FMM-PLS)。该方法:采用减法c-均值聚类进行数据划分,按隶属度最大原则,合理划分子空间,确定予空间个数为3个,然后利用PLS方法:建立3个子模型,并对各子模型的输出进行隶属度加权预测输出值。同时,也建立PLS、QPLS、RBF-PLS单模型,并与提出的FMM-PLS方法:相比较。PLS、QPLS、RBF-PLS和FMM-PLS的最大误差分别为4.9541、4.6282、4.7517、3.8040;均方根误差分别为1.8599、1.7025、1.7381、1.5327。研究结果:表明,与PLS、QPLS、RBF-PLS相比,在航煤干点的估计中本文提出的FMM-PLS方法:预测精度更高,泛化性能更好。Aiming at the difficulty in estimating the Dry Point of Aviation Kerosene Oil in complex conditions, in this paper, a Fuzzy Multi-Model modeling method based on PLS（FMM-PLS） is proposed for estimating the Dry Point of Aviation iKerosene Oil. We use the method of subtractive fuzzy c-means clustering to divide the data, divide reasonable subspaces on the basis of the maximum principle of membership and determine three subspaces, and then used PLS to establish three sub-models. Finally the degrees of membership are used for combing several sub-models to obtain the prediction value. At the same time, we build PLS, QPLS, RBF-PLS sing].e model and make them in contrast with FMM-PLS. The maximum error of PLS, QPLS, RBF-PLS and FMM-PLS are 4.9541, 4.6282, 4.7517, 3.8040; the root mean square error are 1.8599, 1.7025, 1.7381, 1.5327. Through soft-sensing modeling for the Dry Point of Aviation Kerosene Oil of the atmospheric tower, the result shows that prediction accuracy of the FMM-PLS method mentioned in the paper is higher, and it has the better generalization capability compared with the single model method such as PLS, QPLS, RBF-PLS.

关 键 词：PLS 模糊多模型(FMM) 减法c-均值聚类 软测量 

分 类 号：TQ015.9[化学工程] TP391.9[自动化与计算机技术—计算机应用技术]