偏最小二乘法预测头孢菌素C工业发酵过程中发酵液流变特性

Fermentation Broth Rheology Prediction of Industrial Cephalosporin C Process Based on Partial Least Squares Regression

在头孢菌素C工业发酵过程中,发酵液表现出典型的非牛顿流体性质,并可用幂律模型充分描述。传统经验型模型并不能很好地预测头孢菌素C工业发酵过程的流变特性,原因在于模型过于简化,忽略了潜在因素如底物浓度、进料方式、培养基组成等对流变特性的影响,但更多变量的引入会显著增加模型复杂度,且这些参数的动态变化存在相关性,为了解决这个问题,采用了偏最小二乘法进行建模。标准偏回归系数可以量化不同因素对流变参数的作用,其中节孢子比率和菌丝平均分支长度对流变学特性的影响最大。使用偏最小二乘回归(PLSR)模型,可以很好地预测头孢菌素C发酵液的流动行为指数和稠度指数,其相关系数分别为0.94和0.91,具有较好的实用性。

The aim of this study was to quantify the effects of multiple factors on fermentation broth rheology.Industrial fed-batch fermentations of Acremonium chrysogenum were conducted, and rheology properties of samples were adequately described by power law model. Nonlinear modeling taking only fungal morphology and cell concentration into consideration led to poor correlation and little prediction function. One of the reasons probably was that the model was oversimplified and some inconspicuous but significant factors were omitted. In this context, extra elements such as substrate concentration, feed mode, media composition were taken into account, following tremendously increased sample library and existence of variables multicollinearity. Two major morphologies of Acremonium chrysogenum were observed in fermentation broth, i.e., freely dispersed arthrospores and filamentous mycelium. The number of arthrospores was the major factor contributing to rheology properties, based on the standard partial regression coefficients. Using the partial least squares regression(PLSR) model, good prediction of flow index and consistency index can be made from linear recombination of variables, with R^(2)=0.94, R^(2)=0.91 respectively.