基于核偏鲁棒M-回归的间歇反应过程混合模型辨识

Identification of Hybrid Models for Batch Reaction Processes Based on Kernel Partial Robust M-Regression

提出了一种混合模型两步辨识策略,用以解决间歇反应过程的建模问题,并能够有效融合先验知识及过程数据信息。该策略将混合模型的同步辨识分解成为两个独立的步骤,首先确定混合模型的结构,并利用Tikhonov正则化方法实现间歇反应过程反应速率的精确估计;接下来采用核偏鲁棒M-回归(kernel partial robust M-regression,KPRM)算法建立过程变量与反应速率间的经验模型,从而有效抑制过程数据中离群点的影响。利用半间歇过程仿真实验对所提出的策略进行验证,获得了相比于传统方法更高的估计及预测精度。

In order to resolve the problem of modeling batch reaction process, a two-step strategy was presented for the identification of hybrid models, where the prior knowledge and the information provided by process data can be effectively integrated. The simultaneous identification of hybrid model was decomposed into two separate steps by the proposed strategy. At the first step, the structure of hybrid model was determined, and Tikhonov regularization method was employed to estimate the reaction rates of batch reaction process precisely. At the second step, kernel partial robust M-regression （KPRM） algorithm was adopted to calibrate the empirical model between the process variables and the reaction rates, where the negative effects of outliers in process data can be effectively suppressed. A semi-batch simulation experiment was used to verify the proposed strategy, and comparing with traditional method, higher estimation and prediction accuracy were obtained.