NonliNonlinear GPC with In-place Trained RLS-SVM Model for DOC Control in a Fed-batch Bioreactor  

基于同址训练递推最小二乘支持向量机的分批补料生物反应器中溶氧浓度的非线性广义预测控制(英文)

作　　者：冯絮影 于涛 王建林 

机构地区：[1]College of Information Science and Technology,Beijing University of Chemical Technology

出　　处：《Chinese Journal of Chemical Engineering》2012年第5期988-994,共7页中国化学工程学报（英文版）

基　　金：Supported by the National Natural Science Foundation of China (20476007, 20676013)

摘　　要：In this study, Saccharomyces cerevisiae (baker's yeast) was produced in a fed-batch bioreactor at the optimal dissolved oxygen concentration (DOC) and growth medium temperature. However, it is very difficult to control the DOC using conventional controllers because of the poorly understood and constantly changing dynamics of the bioprocess. A generalized predictive controller (GPC) based on a nonlinear autoregressive integrated moving average exogenous (NARIMAX) model is presented to stabilize the DOC by manipulation of air flow rate. The NARIMAX model is built by an improved recursive least-squares support vector machine, which is trained by an in-place computation scheme and avoids the computation of the inverse of a large matrix and memory reallocation. The proposed nonlinear GPC algorithm requires little preliminary knowledge of the fermentation process, and directly obtains the nonlinear model in matrix form by using iterative multiple modeling instead of linearization at each sampling period. By application of an on-line bioreactor control, experimental results demonstrate the robustness, effectiveness and advantages of the new controller.In this study, Saccharomyces cerevisiae （baker＇s yeast） was produced in a fed-batch bioreactor at the optimal dissolved oxygen concentration （DOC） and growth medium temperature. However, it is very difficult to control the DOC using conventional controllers because of the poorly understood and constantly changing dynamics of the bioprocess. A generalized predictive controller （GPC） based on a nonlinear autoregressive integrated moving average exogenous （NARIMAX） model is presented to stabilize the DOC by manipulation of air flow rate. The NARIMAX model is built by an improved recursive least-squares support vector machine, which is trained by an in-place computation scheme and avoids the computation of the inverse of a large matrix and memory reallocation. The proposed nonlinear GPC algorithm requires little preliminary knowledge of the fermentation process, and directly obtains the nonlinear model in matrix form by using iterative multiple modeling instead of linearization at each sampling period. By application of an on-line bioreactor control, experimental results demonstrate the robustness, effectiveness and advantages of the new controller.

关 键 词：nonlinear generalized predictive controller recursive least squares support vector machine in-place computation fed-batch bioreactor dissolved oxygen concentration 

分 类 号：TQ033[化学工程]