基于数据驱动的混杂系统建模与优化控制研究

Method of Hybrid System Modeling and Optimizing Control Based on Data-driven

针对一类非线性、多变量的工业过程,提出了一种基于数据驱动的多输入多输出混杂系统模型辨识方法和基于多参数规划的实时非线性预测控制方法。该方法采用高维数据空间的聚类划分方法,成功实现了MIMO形式的分段仿射含输入自回归(PWARX)模型的辨识;利用混杂系统的HYSDEL建摸语言及其编译器工具,将得到的MIMO-PWARX模型转化成非线性的分段仿射(PWA)模型,采用多参数规划的方法,设计基于PWA模型的预测控制器。该方法实现了从过程数据中挖掘出系统的混杂模型的信息,自动生成复杂过程的PWA模型,并实现了实时的非线性预测控制。最后利用该方法成功实现了四容水箱非线性多输入多输出过程模型的在线控制,得到较好的控制效果。

A multi-input multi-output hybrid system model identification method based on data-driven and a real-time nonlinear predictive control method based on multi-parametric programming were proposed for a class of nonlinear and multivariable industrial processes. A clustering classification method of highdimensional data space was adopted to solve a MIMO piecewise affine autoregressive exogenous （PtVARX） model identification problem. With the help of HYSDEL modeling language of hybrid systems and its compiler tools, MIMO-PWARX model was transformed into nonlinear piecewise affine （PWA） model. Using multi-parametric programming method, a predictive controller based on PWA model was designed to achieve on-line control of nonlinear process. Through this method, the hybrid model information of systems was obtained from a set of process data and PFCA model of complex process was automatically generated. Based on the PWA Process model, real-time nonlinear predictive controller was implemented. Finally, the hybrid identification and control method was applied to the typical MIMO nonlinear quadruple tank system. High identification accuracy and satisfied control performance were obtained.