【摘要】 针对实际工业过程数据往往具有高斯与非高斯共存、非线性的特点，及传统核主成分分析在特征提取时仅考虑变量方差最大化导致信息丢失，并且处理非高斯数据能力欠佳，因此传统核主成分在故障诊断分析中难以取得令人满意的结果的问题，提出一种核熵成分分析(kernel entropy component analysis, KECA)、余弦相似度K均值(K-means of cosine similarity, CSK)的故障诊断方法。首先将数据投影至高维空间，KECA方法根据Renyi熵大小选取投影方向，对数据进行降维；然后，基于KECA能够以角结构方式捕捉数据内部的集群结构信息，设计一种余弦相似度的K均值聚类器，对数据进行聚类并建立故障诊断模型；最后，将KECA-CSK方法应用于田纳西-伊斯曼化工过程进行仿真实验。结果表明：与核主成分聚类分析相比，KECA-CSK方法具有更好的诊断结果，验证了所提方法的优越性。更多还原

【Abstract】 Actual industrial process data often have the characteristics of Gaussian and non-Gaussian coexistence and nonlinearity, and kernel principal component analysis(KPCA)only considers the maximization of variable variance in feature extraction, resulting in information loss, and the ability to process non-Gaussian data is not good, so it is difficult to achieve satisfactory results in fault diagnosis analysis. In view of the above problems, a fault diagnosis method of kernel entropy component analysis(KECA) and K-means of cosine similarity(CSK) was proposed. Firstly, the data was projected into a high-dimensional space to select the projection direction according to the Renyi entropy to reduce the dimension of the data; then, KECA can capture the cluster structure information inside the data in an angular structure, a K-means of cosine similarity was designed to cluster the data and construct the fault diagnosis model. Finally, the KECA-CSK method was applied to the Tennessee-Eastman chemical process for simulation experiments. Simulation results showed that the KECA-CSK method had better diagnostic results than the KPCA, which verified the superiority of the proposed method.更多还原