基于二维模型预测控制的迭代学习控制性能评估方法

Data-driven performance assessment method for iterative learning control based on two-dimensional MPC benchmark

实际工业生产常用操作方式是批次过程,可以用迭代学习控制进行控制。针对过程模型未知的情形,采用子空间辨识求解模型参数,以迭代学习控制的性能评估为课题。首先,根据测量得到的输入输出数据构造Hankel矩阵,再借助QR分解、SVD分解等几何工具计算出子空间矩阵,最后由列子空间与行子空间估计出系统模型参数。在获得模型之后,把迭代学习控制作用下的闭环系统转换成二维模型。接着,提出了基于模型预测控制的基准,并给出了详细的推导证明。根据这些算法,得到最优控制律,并拟合出性能评估曲面,从而性能差距可以更直观地反映在三维空间中。最后,通过仿真对上述方法的效果进行了验证。

Iterative learning control is an extremely effective strategy in batch processes,which are widely used in industrial production.This paper focuses on the performance assessment of iterative learning control.During unknown process model,this study uses subspace identification to estimate model parameters.The algorithm is summarized as follows:Hankel matrices are constructed according to the measured input and output data,and subspace matrices are calculated with geometric tools such as QR decomposition and SVD decomposition,and finally the system model parameters are estimated from the column subspace and row subspace.After obtaining the model,the closed-loop system under iterative learning control is transformed into a two-dimensional model.Afterwards,the 2D model predictive control benchmark is proposed and the detailed derivation and proofs are provided.According to these algorithms,the optimal control principle is derived and a performance assessment surface is fitted.Thus,the performance gap can be more directly reflected in the three-dimensional space.Finally,the results are verified by simulation.