基于LS_SVM的在线建模方法在大肠杆菌发酵过程中的应用  

Online modeling method based on least squares support vector machine for escherichia coli fermentation processes

作　　者：公彦杰[1] 高学金[1] 齐咏生[1,2] 王普[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]内蒙古工业大学信息工程学院,内蒙古呼和浩特010051

出　　处：《计算机与应用化学》2010年第9期1216-1220,共5页Computers and Applied Chemistry

基　　金：国家自然科学基金资助项目(60704036)

摘　　要：为了实现大肠杆菌发酵过程菌体浓度基于模型的优化控制,提出一种基于最小二乘支持向量机(LS_SVM)在线建立大肠杆菌发酵过程局部模型的方法。该方法首先将当前批次滑动时间窗内采集到的数据作为查询序列,以动态时间弯曲距离(DTW)为判断时间序列相似性的标准,从历史批次数据库中搜索与之相似度最高的数据区间,组成训练样本集。然后利用LS_SVM在线建立发酵过程的局部模型。最后通过实际的大肠杆菌发酵过程数据,对本建模方法与传统的LS_SVM离线全局建模方法进行了仿真验证。结果显示,本方法在线建立的大肠杆菌发酵过程菌体浓度预估模型具有更高的精度以及动态适应能力。In order to achieve model-based optimal control of cell concentrations in E.coli fermentation process, a new online modeling method is proposed for modeling Escherichia coli fermentation processes, based on least squares support vector machine（LS_SVM）. In this method, a set of data within the sliding window was set as a query sequence in the current process, and then search for the highest similarity of sample points from the history of batch database to form the training set, based on dynamic time warping（DTW）. At last, this training set will be used for build online local model based on LS_SVM. Based on the proposed modeling method, the predictive models of Escherichia coli concentration are obtained by using the data sampled by the escherlehia coli fermentation processes. The results shows that this method has a higher forecast accuracy and dynamic adaptability compared with the traditional offiine modeling method.

关 键 词：在线建模 最小二乘支持向量机 动态时间弯曲距离 大肠杆菌发酵 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]