基于双层聚类与GSA-LSSVM的汽轮机热耗率多模型预测

针对单模型难以精确描述具有复杂非线性特性的汽轮机热耗率的问题,提出一种新的热耗率多模型建模方法。首先应用GK算法分析出最优聚类个数以及初始聚类中心,避免了聚类数确定的盲目性;然后利用核模糊C均值算法对热耗率样本集做出聚类划分,在每个子空间中利用最小二乘支持向量机（LSSVM）辨识出相应子模型,同时,为了保证子模型精确度,采用引力搜索算法来解决LSSVM参数优化问题;最后,将子模型通过隶属度值加权融合得到精确的热耗率预测模型。以某600MW超临界汽轮机组为研究对象,基于现场数据建立汽轮机热耗率预报模型,仿真结果验证了提出的多模型建模方法具有较高的预报精确度和泛化能力。...

Full description

Saved in:
Bibliographic Details
Published in电机与控制学报 Vol. 20; no. 3; pp. 90 - 95
Main Author 牛培峰 刘超 李国强 张维平 陈科
Format Journal Article
LanguageChinese
Published 国家冷轧板带装备及工艺工程技术研究中心,河北秦皇岛066004%燕山大学工业计算机控制工程河北省重点实验室,河北秦皇岛,066004%秦皇岛职业技术学院机电工程系,河北秦皇岛,066100 2016
燕山大学工业计算机控制工程河北省重点实验室,河北秦皇岛066004
Subjects
多模型
引力搜索算法
最小二乘支持向量机
热耗率
聚类
聚类
multi-model
cluster
引力搜索算法
gravitational search algorithm
最小二乘支持向量机
热耗率
heat rate
least square support vector machine
多模型
Online AccessGet full text
ISSN1007-449X
DOI10.15938/j.emc.2016.03.014

Cover

More Information
Summary:针对单模型难以精确描述具有复杂非线性特性的汽轮机热耗率的问题,提出一种新的热耗率多模型建模方法。首先应用GK算法分析出最优聚类个数以及初始聚类中心,避免了聚类数确定的盲目性;然后利用核模糊C均值算法对热耗率样本集做出聚类划分,在每个子空间中利用最小二乘支持向量机（LSSVM）辨识出相应子模型,同时,为了保证子模型精确度,采用引力搜索算法来解决LSSVM参数优化问题;最后,将子模型通过隶属度值加权融合得到精确的热耗率预测模型。以某600MW超临界汽轮机组为研究对象,基于现场数据建立汽轮机热耗率预报模型,仿真结果验证了提出的多模型建模方法具有较高的预报精确度和泛化能力。
Bibliography:23-1408/TM
multi-model; heat rate; gravitational search algorithm; least square support vector machine; cluster
NIU Pei-feng, LIU Chao , LI Guo-qiang , ZHANG Wei-ping , CHEN Ke（ 1. Key Lab of Industrial Computer Control Engineering of Hebei Province, Yanshan University, Qinhuangdao 066004, China ; 2. National Engineering Research Center for Equipment and Technology of Cold Strip Rolling, Qinhuangdao 066004, China; 3. Department of Electromechanical Engineering, Qinhuangdao Institute of Technology, Qinhuangdao 066100, China）
Aiming at the issue that the characteristic of complex nonlinearity of heat rate for steam turbine which was difficult to be descript accurately by the single-model,a new multi-model modeling method for heat rate was presented. Firstly,the initial cluster centers and optimized cluster numbers were obtained by the G-K algorithm. Then,the data set was clustered into several local regions with kernel-based fuzzy C-means clustering algorithm. In addition,the sub-model was built by least square su
ISSN:1007-449X
DOI:10.15938/j.emc.2016.03.014