基于数据驱动子空间方法的多工况过程预测控制

Multiphase operation mode with transitions from phase to phase is the important characteristics of many nonlinear processes. Batch processes，being an important kind of the multiphase mode process, play an important role in today's industrial manufacturing.The processes exhibit a number of characteristics that lead to interesting control problems. Due to the high dimensionality and complexity of batch processes and the quick product-to-market time required in contemporary industrial settings, it is difficult to create batch process models based on first-principles. Therefore, in this research, the applications of subspace methods, which require only process history data, have attracted research attention in predictive control of batch processes.First, a three-dimensional data matrix should be unfolded to a two-dimensional data matrix or split into several two-dimensional data matrices. The dynamics within each batch can be captured by subspace methods. Based on the dimensionality reduction of batch processes, the complexity of controller design can be relieved to some extent. Then, according to the important multiphase mode characteristic, the proposed research will present a new nonlinear system adaptive model subspace forgetting algorithm. To ensure both quality consistency of the manufactured products and safe operation of this kind of batch process, different multiphase modes divisions are proposed and the method properties are discussed. Finally, the data-driven Model Predictive Control (MPC) algorithms are developed for trajectory tracking and disturbance rejection in batch processes. The algorithms are based on multiphase performance benchmark developed by subspace dynamic matrices and the lustering optimiaztion problems are solved using sensitivity methods.The proposed MPCs can handle constraints. Further-more, due to the batch-wise unfolding approach selected in the data pre-filtering section, the nonlinear time-varying behavior of batch processes is captured thereby yielding very simple and computationally fast data-driven batch MPC.

很多复杂非线性过程具有多工况特性，如间歇过程的过程变量相关关系并非随时间时刻变化，而是呈现出分时段的区域特性，这一特性使得难以把多工况过程当作一个MIMO系统进行建模并据此实现控制器的设计。因此，本项研究面向多工况间歇过程的优化控制需求，应用子空间方法实现对过程时变动态和演变规律的表征，提出数据驱动的多工况过程预测控制理论框架，实现对控制器设计复杂度的改善，以及对间歇过程控制精度的提高。包括如下内容：(1)基于子空间方法提出高维数据的展开预处理方式，支持对分析单元进行时间和批次的双向扩展，并实现对多工况间歇过程的降维处理；(2) 提出带区域遗忘因子的子空间方法，将过程分解成若干操作子时段，建立起基于多工况的过程分析和优化方法；(3)设计多工况过程的性能评价基准，基于灵敏度方法进行分片优化问题的求解，改善对间歇过程的跟踪和抗干扰控制能力，实现数据驱动的多工况过程预测控制器设计。

在很多复杂的非线性过程中，过程变量相关关系呈现出明显的多工况特性，过程变量之间的预测关系并不是随着操作时间时刻变化，而是具有明显的分时段和层级特性，这些特性决定了该类非线性过程的建模与辨识十分复杂。因此，本项研究面向这一非线性过程动态特性的表征需求，深入分析每一子时段的潜在过程相关特性，针对具有多尺度、间歇性、分级等复杂非线性特性的海洋湍流过程展开研究，包括如下研究内容：(1) 针对南海长时间海洋湍流观测数据集，基于子空间方法和奇异谱分析方法，提取数据 Hankle矩阵特征值等关键特征，描述南海湍流在不同尺度下的时间演化规律；(2) 在Block分析建模分时段区域特性的基础上，基于EMD本征函数分解表征非线性湍流能量传递过程的多尺度结构，给出海洋湍流能量级串过程的定量化描述；(3) 针对多时段过程在线监测需求，基于新的时段划分和数据正规化展开方法，提出海洋滑翔机湍流检测的控制性能监测和评价准则，并基于降维数据Hankel矩阵，不经过系统辨识步骤，直接获得无约束系统的解析控制律描述，实现数据驱动的预测控制器设计。