基于子空间辨识的模型预测控制器经济性能评估方法  

A subspace-identification based economic performance assessment approach of model predictive control

作　　者：罗芝芬[1] 田学民[1] 

机构地区：[1]中国石油大学(华东)信息与控制工程学院,山东东营257061

出　　处：《计算机与应用化学》2011年第7期863-866,共4页Computers and Applied Chemistry

基　　金：国家863计划项目(编号:2007AA04Z193);山东省自然科学基金资助项目(编号:Y2007G49)

摘　　要：目前模型预测控制器(MPC)的经济性能评估(EPA)方法多依赖于过程解析的稳态模型,为了更方便有效的对MPC经济性能进行评估,本文提出一种基于子空间辨识的模型预测控制器经济性能评估方法,仅利用过程操作数据即可实现对控制器的经济性能评估,避免了对过程解析稳态模型的依赖性。利用已知过程操作数据,通过子空间辨识求取子空间矩阵,进而得到各输入输出变量间的稳态关系及LQG控制策略下各输入输出变量方差与加权系数的关系并将其直接应用于控制器经济性能优化。在Woodberry塔上的仿真实验验证了该方法的可行性和有效性。The Economic Performance Assessment （EPA） approaches of Model Predictive Control（MPC） mostly rely on process resolved steady-state model. In order for more convenient and effective economic performance assessment, this paper proposed a method based on subspace-idenfificatiun. Only the process operating data is needed to realize the EPA of controllers, avoiding dependence on process resolved steady-state model. The available operating data is used to obtain the subspace matrix through^subspace identification. Then the steady-state relationship between input and output variables and the relationship between their variances and weighting coefficients under LQG control strategy are achieved for direct application on the economic performance optimizing. Simulation results on Woodberry show that the approach has good feasibility and effectiveness.

关 键 词：模型预测控制器 经济性能评估 子空间辨识 LQG基准 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置] TP319[自动化与计算机技术—控制科学与工程]