Abstract:

The penicillin fermentation process is a typical nonlinear, dynamic, multiphase, and uncertain process. A single model-based soft sensor is difficult to meet the requirements of system for estimation accuracy. A multi-model soft sensor method based on the improved density peak clustering algorithm has been proposed in this paper to estimate the product concentration in penicillin fermentation process. Firstly, a similarity function instead of the Euclidean distance is introduced to calculate the k-nearest neighbors of the sample points, and the shared neighbors between the sample points and their k-nearest neighbors are computed, then the k-nearest neighbors and the shared neighbors are used to redefine the local density for the sample points. Secondly, the k-nearest neighbors between sample points is used to redefine the allocation strategy of sample points. Finally, the improved clustering algorithm is used to obtain clustering subsets, and the soft sensors based on least squares support vector machine for each subset are established. The verification results of the Pensim simulation platform show that the improved clustering algorithm can more accurately cluster the sample data, thereby effectively improving the estimation accuracy of the soft sensor model of the penicillin fermentation process.

Key words: improved density peak clustering, algorithm, soft sensor, model, fermentation

摘要：

青霉素发酵过程具有较强的非线性、时变性、阶段性和不确定性，基于单一的软测量模型对产物浓度进行在线估计，难以满足系统对模型精度的要求。针对上述问题，提出一种改进密度峰值聚类的多模型软测量建模方法来估计青霉素发酵过程中的产物浓度。首先，引入相似度函数代替欧氏距离计算样本点的k近邻，并且计算样本点与其k近邻之间的共享近邻，进而利用样本点的k近邻及共享近邻重新定义样本点的局部密度。其次，利用样本点之间的k近邻关系来重新定义样本点的分配策略；通过改进的聚类算法得到各聚类子集，分别建立基于最小二乘支持向量机的软测量模型。Pensim仿真平台的验证结果表明，改进的聚类算法能够更加准确地对样本数据进行聚类，从而有效提高青霉素发酵过程软测量模型的估计精度。

关键词: 改进密度峰聚类, 算法, 软测量, 模型, 发酵