基于KPLS数据重构的非线性过程监测与故障辨识

为适应快速变化的化工产品需求，过程工业逐步向柔性生产发展，使得间歇过程的应用日益广泛。这一类工艺过程具有动态和非线性的特征，过程故障带来的工艺波动和安全风险是较为突出的挑战。采用基于核函数的偏最小二乘方法，在高维特征空间提取特征变量，这些变量包含了生产过程的非线性结构特征，也反应了过程工况的模式特征。针对传统线性方法存在的故障漏报等问题，利用核函数技巧，在特征空间进行数据重构，进而计算统计监控指标SPE，并通过对SPE的在线监测实现更加有效地故障辨识。本方法针对标准非线性测试对象进行了过程监测，实现结果充分说明了方法的有效性。

Nonlinear process monitoring and fault detection based on data reconstruction using KPLS

In this paper, a new nonlinear process monitoring technique based on kernel partial least squares (KPLS) is developed. KPLS has emerged in recent years as a promising regression method for tackling nonlinear systems, such as batch processes. KPLS can effectively capture the nonlinear relationship in the process variables and is potentially suitable for monitoring nonlinear process disturbances. To resolve the fault missing problem of the traditional linear methods, a simple approach of calculating the squared prediction error (SPE) based on data reconstruction in the feature space is suggested. Based on SPE charts in the feature space, KPLS was applied to fault detection in Tennessee Eastman benchmark process. The proposed approach showed superior performance in process monitoring and quality variables prediction compared to the conventional PLS methods.