中文摘要：

针对蒸汽裂解实验装置的开工过程具有间歇操作，变量间相关性高的特点，传统的故障识别方法无法有效处理这种具有较强动态特性的实际工业生产过程。本文提出利用主元分析，用少量主元反映多个动态变量的综合信息，并利用正交小波变换的多尺度时频分析提取主元中表征工况变化的频带特征，对频带特征进行模式归纳分类，进而识别工况。实验结果证实了所提出方法的可行性和有效性。

英文摘要：

The start-up process of steam cracking experimental device had the characteristics of the intermittent operation and high correlation of variables. Meanwhile, the traditional fault identification methods can not effectively deal with such a strong dynamic characteristics of the actual industrial production process. The paper proposed to use PCA, which can make principal components with a small number of dynamic variables reflect the number of comprehensive information, then using the time-frequency analysis of multl-scale of the orthogonal wavelet transform to extract the frequency bands. The characteristics of the frequency bands which performance the change in the principal component of the status, and the band features categorized by model, further more identifying the status. The experimental results show that the proposed method is feasible and effective.