基于改进FCM聚类算法的混合建模方法在苯酚浓度预测中的应用  

Application of Hybrid Modeling Method in Phenol Concentration Prediction Based on Improved FCM Clustering Algorithm

作　　者：周达左[1] 陶洪峰 ZHOU Da-zuo;TAO Hong-feng(College of Mechanical and Electrical Engineering,Changzhou Vocational Institute of Textile and Garment;MOE Key Laboratory for Advanced Process Control of Light Industry,Jiangnan University)

机构地区：[1]常州纺织服装职业技术学院机电学院 [2]江南大学轻工过程先进控制重点实验室

出　　处：《化工自动化及仪表》2023年第6期889-892,899,共5页Control and Instruments in Chemical Industry

基　　金：常州纺织服装职业技术学院资助项目(批准号:51800222107)资助的课题。

摘　　要：为了解决单一模型无法满足复杂化工生产过程预测精度要求的问题,引入混合建模方法。首先,考虑到模糊C均值聚类(FCM)算法在初始聚类中心选择上存在的缺陷,采用SA算法和GA算法对其进行优化,以选择最合适的初始聚类中心,提高聚类精度;然后,基于支持向量机建立各子类预测模型;最后,将测试样本划分到各子类中,采用各子类模型仿真得到预测值。采用混合建模方法和单模型方法预测苯酚浓度并与真实值对比,结果表明:笔者所提混合模型得到的平均相对误差(MRE)和最大相对误差(MXRE)均小于单模型的。Considering the fact that a single model fails to meet the precise prediction of complex chemical production,a hybrid modeling method was introduced.Due to the fuzzy C-means clustering(FCM)algorithm’s shortcoming in selecting the initial clustering center,both SA algorithm and GA algorithm were adopted to optimize it so as to select a proper initial clustering center and improve the clustering accuracy,including having the support vector machine(SVM)based to establish the prediction model of each subclass,and having the test samples divided into sub-classes,and having the predicted values obtained by simulating each sub-class model.Compared the true value with phenol concentration predicted by the hybrid modeling method and the single model method shows that,the mean relative error(MRE)and maximum relative error(MXRE)obtained by the proposed hybrid model are smaller than those obtained by the single model.

关 键 词：混合建模 改进FCM聚类算法 支持向量机 相对误差 苯酚浓度 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]