集成最大均值差异正则约束的迁移子空间软测量

Transfer Subspace Soft Sensor with Maximum Mean Discrepancy and Regularization Constraints

针对流程工业中湿式球磨机工况变化后,数据分布差异导致的原测量模型失准问题,引入集成最大均值差异、正则约束的迁移子空间(the transfer subspace with integrated maximum mean discrepancy and regular constraint MRTS)软测量建模方法。该方法首先在源领域和目标领域上训练2个耦合投影矩阵,将源域和目标域映射到2个低维子空间中,然后,集成最大方差、最大均值差异及正则项,通过优化求解得到这2个特征耦合的变化矩阵,最后在源领域所构建的子空间中建立回归模型。实验室磨机负荷参数的预测结果表明,该方法优于传统软测量建模方法,能够有效提高模型的预测精度,对实际流程工业具有一定的指导意义。

Aiming at the problem that inconsistent original model caused by the distribution discrepancy,when the working conditions are changed in the process industry, the soft sensor algorithm of the transfer subspace with integrated maximum mean discrepancy and regular constraint is introduced. The method first learns two coupled projection matrices on the source and target domains,which maps the two domains into a low-dimensional subspaces,and then integrates the maximum variance,the maximum mean discrepancy,and the regular terms,to find the two features,which are coupled to the projection matrix,and finally a regression model is established in the subspace to obtain the predict value of the mill load parameters. The prediction of the laboratory mill load parameters shows that the method is superior to the traditional soft sensor methods and can effectively improve the prediction accuracy of the model,which has certain guiding significance for the actual process industry.