模拟移动床过程的软测量建模仿真研究

Simulation research on soft-sensor modeling of Simulated Moving Bed process

模拟移动床色谱作为主要的现代吸附分离技术,近年来在石油化工、精细化工、生物制药、食品加工等领域的复杂混合物分离过程中得到越来越多的应用。模拟移动床系统是一种具有强非线性、强耦合、混杂性、分布参数等特征的复杂工程系统,其建模问题一直受到广泛关注。本文提出用神经网络与遗传算法结合的混合建模方法建立SMB色谱分离过程4区流量与组份纯度的软测量模型。为了解决RBF神经网络训练时隐含层节点数选取无依据,只能依靠反复仿真尝试的问题,提出将混合递阶遗传算法与RBF神经网络结合,建立SMB色谱分离过程软测量模型,以较高的精度实现了SMB组份纯度软测量,通过仿真验证了方法的可行性。

The Simulated Moving Bed （SMB） chromatography has been used extensively for the complex separation tasks in the areas of petroleum chemicals, fine chemicals, pharrnaeeuticals, and food processing in recent years as a main modem separation technology. As a kind of strong nonlinear, strong coupling, hybrid, distributed parameter characteristic of complex engineering system, process modeling research on simulated moving bed system has been widely concerned. This paper presents a method of combining Neural Networks and Genetic Algorithms to establish the soft sensor model reflecting the relationship between the four zones flow and the component purity of the SMB chromatographic process. In order to avoid the trial and error practice in the selection of hidden layer units of RBF networks training, it is proposed to set up the combine the hybrid hierarchical Genetic Algorithms and RBF Networks in the establishment of an accurate soft sensor model of the SMB chromatographic separation process.The validity of the modeling strategy is confirmed through simulations.