基于样本稀疏化高斯过程的发酵过程软测量建模方法

Soft-sensor modeling method in a fermentation process based on the samples of a sparse Gaussian process

提出了一种基于样本稀疏化高斯过程(GP)的发酵过程软测量建模方法。该方法将聚类和灰色关联度分析相融合,综合考虑样本点间欧式距离和各个特征向量对样本点间相似度的影响,通过剔除相似度比较大的样本点,实现训练样本集的稀疏化,降低了模型的计算复杂度。利用基于样本稀疏化的高斯过程构建青霉素发酵过程的软测量模型,同时得到青霉素浓度的预估值和表征预估值的不确定度,实验结果表明,本文所提方法与标准GP方法相比,在保证模型预测精度的前提下,减少了模型的训练时间。

A soft-sensor model method has been proposed for a fermentation process based on the samples of a sparse Gaussian process( GP). A method based on clustering and grey correlation analysis to select effective sample subsets was employed. This incorporated the Euclidean distance between sample points and the feature vector similarity between the sample points,whilst eliminating the sample points which had larger similarity. The method can give not only the forecast value of the fermentation,but also the forecast precision of the model. The experimental results show that the soft-sensor model method of a sample sparse GP can not only guarantee the accuracy of predictions but also save on training time of the model.