针对化工连续生产过程的时序性及非线性等特征,提出一种新的基于数据驱动的化工过程故障诊断方法:ISOMAP-LDA。首先实行流形学习算法ISOMAP,在保持量测数据几何结构特性下完成非线性降维,然后基于提取的嵌入变量张成的低维空间,选用线性判别分析(LDA)构造故障模式类的判别函数,负责各采样个体故障类型的判定。将该方法用于仿真化工Tennessee East man过程的故障诊断,结果表明,ISOMAP-LDA方法不仅拥有较高的故障诊断能力,而且取得采样在低维空间的可视化表示。
A new method by integrating the multivariate statistical analysis with neural network used for complex pattern classification was proposed in this paper. First, a particularly developed statistical method called correlational components analysis was employed to extract pattern characteristics from the original sample pattern space. These pattern characteristics were then used as inputs to a multi-layered feedforward neural networks for further pattern classification, The proposed approach transforms the complex patterns into lower dimensional and mutually decoupled ones, it also takes the advantages of the self-learning capability of the neural networks. Finally, a practical example of natural spearmint oil was used to verify the effectiveness of the new method. The results showed that the proposed integrated approach gives better results than other conventional methods.
