基于鲁棒核偏最小二乘法的化工过程故障检测

Fault detection for chemical process based on robust KPLS

针对传统核偏最小二乘法不能对包含离群点的非线性数据进行有效建模的问题,提出了鲁棒核偏最小二乘(R-KPLS)算法。该算法先利用鲁棒标准化代替传统的标准化,然后利用加权策略与传统的核偏最小二乘法结合,实现鲁棒建模。TE数据仿真实验结果表明,R-KPLS算法比传统KPLS算法在处理含有离群点的数据时具有更好的检测性能。

Aiming at the problem that the traditional kernel partial least squares method can’t effectively model the nonlinear data including outliers, a robust kernel partial least squares(R-KPLS) algorithm is proposed. The algorithm first uses robust standardization instead of traditional standardization, and then uses the weighting strategy combined with the traditional kernel partial least squares method to achieve robust modeling. The simulation results of TE data show that the R-KPLS algorithm has better detection performance than the traditional KPLS algorithm in processing data with outliers.