一种发酵过程的分阶段融合建模算法研究

Fusion modeling for fermentation process based on multiple stages

针对发酵过程由于外界因素或者反应进程的改变往往具有多阶段性,不同操作阶段动态特性不同,线性化程度也不同,提出一种分阶段多模型融合算法。首先利用模糊C聚类算法将发酵过程分为多个阶段;然后基于量子粒子群寻优的最小二乘支持向量机建立各个阶段对应的子模型;最后根据预测样本到聚类中心的模糊隶属度作为加权系数,将得到的子模型融合输出。文中利用Pensim仿真平台数据,将提出的算法应用到青霉素发酵过程的软测量建模过程,结果表明此方法具有较高的预测精度和良好的泛化能力。

Due to the external factors or the changes of the reaction process, which have the characteristic of multi-stage, different operating points, dynamics and nonlinearity, a multiple model approach under the framework of LS-SVM is proposed. The FCM algorithm is used for the data preprocessing and gets several stages. Then establish multi-stage least squares support vector machine sub-models which optimized by quantum particle swarm optimization. Finally, the estimation of every sub-model is fused by computing the weighting coefficient which is the fuzzy membership of predict sample to the cluster center. The proposed modeling method is used to construct a soft sensor model for a penicillin fermentation process by using the simulation data from the Pensim simulation platform. The results show that this method has higher prediction accuracy and better generalization ability.