基于MKECA间歇过程多阶段故障监测方法研究

Research on multistage fault monitoring based on MKECA batch process

针对间歇过程的非线性、多阶段性等特点,提出一种基于核熵成分分析（KECA）的间歇过程多阶段故障监测方法。首先将过程数据通过KECA核映射到高维特征空间内,依据核熵与角结构相似度对间歇过程进行阶段划分;接着引入沿批次-变量的三维数据向二维数据展开方式,并在每个子阶段建立多向核熵成分分析（MKECA）非线性故障监测离线模型,采用新型基于角结构相似度计算统计量控制限,无需假设过程变量服从高斯分布;最后,计算监测采样点角结构相似度统计量实现间歇过程的多阶段在线故障监测。本文利用青霉素仿真实验数据进行仿真实验,验证了该方法的可行性与有效性。

In view of the non-linear and time-varying characteristics of intermittent process, a multi-stage fault monitoring method based on kernel entropy component analysis（KECA） is proposed. Firstly, the normal three-dimensional data is expanded into two-dimensional data along the sampling points into the KECA cluster model, and the data are divided into stages in the high-dimensional feature space based on the nuclear entropy and angle structure information. For the traditional on-line monitoring along the batch expansion mode, the shortcomings of the future value of the batch are evaluated and the two-dimensional data along the batch-variable are introduced. The multi-directional kernel entropy component analysis（MKECA） nonlinear fault monitoring model is further established in each sub-stage, and a new type of structure-Multi-stage online fault monitoring of intermittent processes. Finally, the experimental data of penicillin were used to simulate the experiment to verify the feasibility and effectiveness of the method.