基于测地线流式核的隐空间多工况软测量建模

Latent space soft measurement modeling based on geodesic flow kernel in multiple working conditions

针对工业过程中工况变化后的新工况缺乏标记样本构建预测模型,原有软测量模型预测失准的问题,研究了一种基于迁移潜在特征学习的多工况软测量方法。首先,利用测地线流式核框架对多工况下分布不匹配的过程数据进行域适应;然后,通过特征迁移后的历史工况样本与标签变量获取映射矩阵,将多工况迁移样本投影至隐空间;最后,利用支持向量机(SVM)建模。通过田纳西伊斯曼(TE)过程仿真实验的标签预测结果表明本文方法的有效性。

Aiming at the problem that new working condition lack of labeled sample to construct prediction model after working condition changes in industry process,which lead to inaccurate prediction of original soft measurement model,a multiple working condition soft measurement method based on transfer latent feature learning is studied.Firstly,process data of mismatched distribution are conducted for domain adaptation in multiple working condition by using geodesic flow kernel framework.Then,mapping matrix is obtained through historical working condition samples and label variables after feature transfer,and multiple working condition transfer sample is projected to latent space.Finally,support vector machine(SVM)is used for modeling.Effectiveness of the proposed method is verified by label prediction result of Tennessee Eastman(TE)process simulation experiment.