间歇反应过程基于二维系统理论的迭代学习预测控制研究

It is of great significance to develop the advanced control technologies of product quality of batch processes for improving the level of production automation. To address the problems of complex mechanism, diverse interference factors and difficult quality control in the batch processes, iterative learning predictive control (ILPC) based on the linear time-varying perturbation (LTVP) model is theoretically studied by using two-dimensional system theory in this project. The batch processes have the feature of two-dimensional system in both batch and time domains, but traditional iterative learning control is studied only in the batch domain. From the perspective of the dynamic responses described by Roesser model in the two-dimensional system, the evolution rules of system states and tracking errors can be derived theoretically and the convergence conditions of the ILPC method can be revealed more clearly. Furthermore, the uniform framework of combining iterative learning control with predictive control is proposed, and the internal relationships among various kinds of iterative learning control methods can be also revealed. To overcome the problems of initial condition errors and reaction parameter fluctuations, the transmission principles of initial condition and reaction parameter are described on the basis of the dynamic responses of the two-dimensional system, and then robust iterative learning predictive control (RILPC) can be developed. To address the problems of finite operation data from the product quality and tracking quality at one and more times, piece-wise linear method is used to build the predictive model of product quality, and then point-to-point iterative learning control of product quality is studied. Finally the systemic theory and methodology of the iterative learning predictive control will be realized in the project.

间歇反应过程的先进控制对于提高生化制药等行业自动化水平具有重要的意义。针对反应过程机理复杂、干扰因素多、质量难控等问题，本项目拟采用二维系统理论研究基于线性时变扰动模型的迭代学习预测控制方法。间歇过程具有时间与批次的二维系统特征，而传统迭代学习控制仅在批次方向分析，本项目将从二维系统Roesser模型描述的系统动态响应出发，用方程解析系统状态演化和跟踪误差传递，从而更清晰地揭示迭代学习预测控制的收敛条件，进而研究预测控制与迭代学习控制两种方法有机结合的统一框架，揭示多种迭代学习控制方法的内在关系。为了克服初始条件偏差、反应参数扰动等问题，分析二维系统动态响应描述下的初始条件和扰动传递过程，研究鲁棒迭代学习预测控制算法。针对反应产物数据有限、单点或多点产品质量跟踪等问题，采用分片线性方法建立产品质量预报模型，研究产品质量的点对点跟踪迭代学习控制方法，形成较系统的迭代学习预测控制理论和方法。

间歇反应过程的先进控制、智能控制对于提高生物制药等行业自动化水平具有重要的意义。本项目针对间歇反应过程机理复杂、干扰因素多、质量难控等问题，研究了基于二维系统理论的预测迭代学习控制方法。间歇过程具有时间与批次的二维系统特征，而传统迭代学习控制仅在批次方向分析，本项目从二维系统Roesser模型描述的系统动态响应出发，首次提出了迭代学习控制的系统状态和跟踪误差的动态全响应解析表达式，解析表达出系统状态的演化和跟踪误差的传递，从而更清晰地揭示了基于2D系统的迭代学习控制方法的响应行为，为分析迭代学习控制方法的收敛性、鲁棒性等问题提供了一种理论基础，具有重要的理论意义。本项目还研究预测控制与迭代学习控制两种方法有机结合的统一框架，并通过2D系统模型揭示了多种迭代学习控制方法之间的内在关系。针对发酵反应结束时刻产品浓度等特定关键点的跟踪问题，本项目研究了基于轨迹更新的点对点综合预测迭代学习控制方法。提出了基于分段输入的点对点综合预测迭代学习控制方法，具有算法简单、计算量较小的特点。针对输入信号受限的情况，提出了结合轨迹更新的点对点综合预测迭代学习控制方法，能够解决实际的批次生产过程中面临的问题，具有较高的实用价值。