基于分阶段的LSSVM发酵过程建模

LSSVM modeling for fermentation process based on dividing stages

发酵过程建模是研究微生物发酵的重要课题,基于模型可实现被测参量的软测量、系统的优化控制。鉴于引入混合核函数的最小二乘支持向量机在过程建模中具有优良表现,采用基于混合核函数的最小二乘支持向量机建模。但由于发酵过程周期较长,最小二乘支持向量机的全局模型预测精度难以保证,算法复杂度很高,因此提出一种分阶段建模方法。首先,选择表征阶段特性的辅助变量,利用模糊C均值聚类算法对样本数据聚类,将发酵过程分成不同的阶段,然后为各个阶段分别建立最优混合核最小二乘支持向量机局部模型,最后将局部模型合成构成过程的完整模型。将此方法应用于青霉素发酵过程和重组大肠杆菌发酵过程中,验证了该方法的有效性。

The modeling for fermentation process is an important issue in studying microbial fermentation.A precise mathematics model lays the foundation for the subsequent soft measurement and optimization work.The least squares support vector machine（LSSVM）with mixtures of kernels was used for its good performance in the modeling for fermentation process.Global modeling treats entire fermentation process as a study object.It is difficult to guarantee the prediction accuracy.Moreover,it brings about high computing complexity.To solve these problems,a staged modeling method based on LSSVM with optimal mixtures of kernels was proposed.Firstly,secondary variables which presented the characteristics of each stage were selected and modeling samples were classified according to the disparate stages by FCM clustering algorithm.Secondly,local models of stages were built using LSSVM while each local model used the optimal mixtures of kernels.Lastly,the integral process model was constituted with these local models.This method was applied to modeling the penicillin fermentation process based on Pensim and the recombinant E.coli fermentation process in the interleukin-2 production.Simulation results showed that local models of stages were superior to global model both in prediction accuracy and time complexity.The effectiveness of the proposed approach was validated.