基于LS-CSO优化MWLS-SVM的SO_(2)排放浓度软测量建模  

SOFT SENSOR MODELING OF SO_(2) EMISSION CONCENTRATION BASED ON MWLS-SVM OPTIMIZED BY LS-CSO

作　　者：梁伟灿 周宾[1] 王汗青 Liang Weican;Zhou Bin;Wang Hanqing(School of Energy and Environment,Southeast University,Nanjing 210000,Jiangsu,China)

机构地区：[1]东南大学能源与环境学院,江苏南京210000

出　　处：《计算机应用与软件》2022年第5期60-67,共8页Computer Applications and Software

基　　金：国家重点研发计划项目(2017YFB0603204);国家自然科学基金青年科学基金项目(50906013)。

摘　　要：针对石灰石-石膏湿法烟气脱硫过程最小二乘支持向量机(LS-SVM)建模中存在的噪声和辅助变量差异性对模型的影响,提出一种混合加权最小二乘支持向量机(MWLS-SVM)的建模方法。该建模方法对辅助变量进行加权,以降低各辅助变量对模型的差异性影响,并采用样本局部异常因子,对模型经验风险项进行加权处理,以提高模型的逼近和泛化能力,同时提出一种局部搜索竞争粒子群(LS-CSO)算法优化MWLS-SVM模型参数和特征权重。分别采用直接加权最小二乘支持向量机(DWLS-SVM)、LS-SVM和MWLS-SVM三种方法对湿法烟气脱硫SO_(2)排放浓度建立软测量模型,采用LS-CSO优化模型参数,实验结果表明MWLS-SVM模型在预测精度和泛化能力表现上明显优于LS-SVM和DWLS-SVM。Aimed at the effects of noise and auxiliary variable differences in the modeling of the least square support vector machine(LS-SVM)in the limestone-gypsum wet flue gas desulfurization process,a mixed weighted least square support vector machine(MWLS-SVM)is proposed.This modeling method weighted the input features to reduce the impact of the auxiliary variables on the model.It used sample local abnormal factors to weight the model’s empirical risk terms to improve the model’s approximation and generalization capabilities.A local search competitive particle swarm optimization(LS-CSO)was proposed to optimize MWLS-SVM model parameters and feature weights.The direct weighted least squares support vector machine(DWLS-SVM),LS-SVM and MWLS-SVM were used to establish soft-sensing models for SO_(2)emission concentrations in wet flue gas desulfurization,and LS-CSO was used to optimize model parameters.The results show that the MWLS-SVM model is significantly better than LS-SVM and DWLS-SVM in terms of prediction accuracy and generalization ability.

关 键 词：最小二乘支持向量机 竞争粒子群 参数优化 特征加权 湿法烟气脱硫 SO_(2)排放模型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]