基于动态PLS框架的多变量无静差预测控制  

Multivariable offset free model predictive control in dynamic PLS framework

作　　者：金鑫[1] 梁军[1] 

机构地区：[1]浙江大学控制科学与工程学院,浙江杭州310027

出　　处：《浙江大学学报（工学版）》2016年第4期750-758,共9页Journal of Zhejiang University：Engineering Science

基　　金：国家自然科学基金资助项目(61174114;U1509203);教育部高校博士点基金优先领域资助项目(20120101130016);浙江省公益性技术应用研究计划资助项目(2014C31019)

摘　　要：针对动态偏最小二乘(DyPLS)建模方法容易导致模型与实际系统的失配,使控制系统产生静差的问题,提出基于DyPLS框架的多变量系统无静差模型预测控制(MPC)方法.将基于状态空间模型的MPC方法推广到DyPLS框架下.在内模型中采用状态空间模型描述系统的动态过程,利用该模型设计MPC控制器.该方法将内模型的状态作为控制系统的反馈,由于模型的失配导致该状态不能准确地描述系统的实际状态,导致了静差的存在.对该框架下的状态空间模型进行增广,引入扰动模型,利用状态观测器估计系统输出与模型输出的偏差.给出该增广模型的能观测性条件.为了采用卡尔曼滤波器的方法求取观测器的增益矩阵,分析原空间的数据投影到潜变量空间后变量方差的变化情况.该方法在控制中引入了输出反馈,保证了控制的无静差跟踪特性,能够抑制系统中的不可测扰动.Jerome-Ray的精馏塔模型的仿真结果验证了该方法的有效性.An offset free model predictive control（MPC）in the DyPLS framework was proposed in order to deal with the problem of system offset which was caused by mismatch between actual plant and dynamic partial least square（DyPLS）model.State space MPC was extended into the DyPLS framework.The state space model was used in inner model to describe the system dynamic,and state space MPC controller was designed.The state of inner model was as the feedback of the control system.This state can＇t describe the real system state and results in offset due to model/plant mismatch.A disturbance model was introduced in the inner state space model in order to solve the problem.A state observer was used to estimate the error between the inner model and system output.Observability condition of augmented model was given.The covariance that the original space data were projected to the latent variable space was analyzed in order to calculate the gain matrix of the observer with Kalman filter.The method used the system output as feedback in the control scheme.The offset free tracking was guaranteed and unmeasured step disturbance can be rejected.The simulation results based on Jerome-Ray distillation column model demonstrated the effectiveness of proposed method.

关 键 词：偏最小二乘 模型预测控制(MPC) 状态空间模型 无静差控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]