中文摘要：

针对复杂化工过程具有的非线性、非高斯性和动态特征，提出了基于核独立成分分析（KICA）的模式匹配方法，用于动态过程监控和诊断。首先，利用滑动窗建立基准集与测试集的KICA模型，提取各自的核独立元；其次，融合余弦函数绝对值度量和距离度量，提出新的不相似度监控指标，识别训练与测试操作期间的相似模式，进行故障检测；最后，基于两类数据的核子空间之间的差异子空间，获得每个过程变量方向与该差异子空间之间的互信息，并定义新的非线性非高斯贡献度指标，进行故障诊断。基于污水处理过程的仿真结果表明，与主成分分析不相似度因子的方法、标准的独立成分分析（ICA）统计指标方法及标准的ICAT^2／SPE指标融合的贡献度方法相比，本文提出的方法具有更好的检测能力与故障诊断效果。

英文摘要：

Aiming at the complex chemical process with strong nonlinearity, non-Gaussianity and dynamics, a new pattern matching approach based on the kernel independent component analysis （KICA） is proposed for process monitoring and fault diagnosis. Firstly, the KICA models are constructed on the normal benchmark and monitored set through the sliding windows strategy. Secondly, a new dissimilarity index is defined, which combines measurements of the cosine absolute value and distance, to evaluate the similarity degree between the benchmark and monitored subspaces for fault detection. Finally, based on a difference subspace between the two subspaces of two patterns, the mutual information between each process variable direction and the difference subspace is derived. Meanwhile, a new nonlinear and non-Ganssian contribution index is defined for fault diagnosis. The simulation results based on the waste water treatment process indicate that the proposed approach provides the better detection capability and diagnosis performance than the principal component analysis based dissimilarity factor, the standard ICA statistic index and the standard ICA T^2/SPE combined contribution method.