基于DKPLS的非线性过程故障检测

DKPLS based fault detection for nonlinear processes

提出了一种基于动态核偏最小二乘(DKPLS)的多变量监测方法.KPLS结合PLS和核技巧两者的优点,通过非线性映射在高维的特征空间中计算出得分向量.所提出的DKPLS将KPLS与系统的动态特性相结合,不仅可以从观测变量的动态信息中取出了那些与输出变量相关的非线性信息,并且还具有动态的优点,由于不考虑动态优化过程,因此它又是一种无教师指导学习方法.计算结果表明DKPLS能有效地提取时间和批次的信息,对于动态过程的故障诊断有很好的效果.

A novel multivariate monitoring approach based on dynamic kernel partial least squares(DKPLS) with dynamical characteristics is proposed.Kernel partial least squares(KPLS) is a nonlinear method for tackling nonlinear processes because it can efficiently compute scores in high-dimensional feature spaces by using nonlinear kernel functions.So DKPLS has the advantages in both KPLS and dynamic characteristic.DKPLS extracts the dynamic information from the observed data,and it does not consider the dynamic optimize process,so DKPLS is an unsupervised method.DKPLS can effectively extract both the time and batch information,and it is better for the fault detection of the dynamic process.The computation results can prove the conclusion and show the effectiveness of the proposed method.