小样本下基于领域自适应的间歇过程质量预测

Quality prediction of batch process based on domain adaptation under few shot

针对间歇过程因样本量不足导致预测精度低的问题,提出一种基于领域自适应的间歇过程质量预测模型方法。首先,引入空间注意力机制自适应地增强与质量指标相关性高的输入变量,结合一维卷积层和长短期记忆网络单元分别挖掘数据的空间特征和时间特征。其次,将领域自适应方法引入到建模过程中,进行目标域数据和源域数据之间特征的自适应匹配,降低2个数据集因分布差异对模型预测精度的影响。该方法在慢时变的青霉素生产过程仿真数据集和酚醛树脂工业生产过程进行了验证。实验结果表明,所提出的模型能有效地提高小样本下间歇过程质量预测的精度。

In batch process modeling,only few samples are collected which is not enough to develop high accurate model.To solve this problem,a model based on domain adaptation under small sample was pro-posed.Spatial attention mechanism was introduced to the model first in order to enhance input variables with high correlation to quality index.After that,one-dimension convolutional layer and long and short term memory units was also borrowed to mine spatial and temporal characteristics of data,respectively.Secondly,the domain adaptation method was introduced into the modeling process to perform adaptive matching of features between the target domain data and the source domain data to reduce the impact of the difference in distribution of the two data sets on the prediction accuracy of the model.The method was validated on the slowly time-varying penicillin production process simulation data set and the phenolic res-in industrial production process.Experimental results showed that the proposed model can effectively im-prove the accuracy of batch process quality prediction under few shot.