基于变分贝叶斯算法的青霉素发酵过程建模

Modeling for penicillin fermentation processes based on variational Bayesian algorithm

摘要:

青霉素发酵过程具有明显的阶段特征，同时由于操作条件多变、生产环境复杂等原因导致其存在极大的不确定性，故本文在变分贝叶斯框架下建立了青霉素浓度预测的FIR融合模型。首先选取调度变量对发酵阶段进行划分，然后基于变分贝叶斯算法辨识得到各FIR子模型的参数，最后根据阶段特征计算样本隶属于各子模型的概率并融合子模型的输出得到青霉素浓度的预测值。文中利用工业规模青霉素发酵罐的实际数据进行仿真实验，模型预测青霉素浓度的相关误差为0.24%，表明提出模型具有较高的拟合度，能够更为精准的预测青霉素浓度并适应实际的复杂工业环境。

Abstract:

Penicillin fermentation processes have obvious stage characteristics, meanwhile which have great uncertainties due to some reasons of variable operation conditions and complex production environments, this paper aims to establish a finite impulse response (FIR) fusion model under the variational Bayesian (VB) framework for online prediction of penicillin concentration. First, the scheduling variable is selected to divide fermentation stages, then the parameters of each FIR sub-model are identified based on the VB algorithm. Finally, the probability of the sample belonging to each sub-model is calculated according to the stage characteristics, and further applied to fuse sub-model outputs for obtaining the penicillin concentration predictions. The paper uses the actual industrial scale penicillin fermentation data to carry out simulation experiments. The correlation error of the model predicting penicillin concentration is 0.24% which shows that the model has a high degree of fitting, which can provide more accurate prediction of the penicillin concentration and adapt to the actual complex industrial environments.