基于改进动态等距离映射的非线性动态故障诊断方法

Nonlinear Dynamic Fault Diagnosis Method Based on Improved Isometric Mapping

分析动态等距离映射算法,针对数据稀疏分布造成短路边的缺点,运用主成分分析法进行可视化一维主元提取,近似确定高维采样点的分布情况,自适应获取采样点的近邻参数.其次,采用流形距离量度代替欧氏距离进一步得到测地线距离,提取训练样本的子流形特征,并运用标准化监控统计量实施过程监控和故障检测.最后,设立子流形综合相似度指标,对故障数据进行模式匹配.TE(Tennessee Eastman)过程的仿真结果表明:所提出的方法可以更为有效地检测到故障发生,并进一步对发生的故障进行识别.

DISOMAP（dynamic isometric mapping） algorithm is analyzed in this paper.According to the shortcomings of short edge caused by data sparse distribution,PCA（principal component analysis） algorithm was used to extract one-dimensional visualization principal component,determining the distribution of sampling points approximately,acquiring the neighbor parameter of sampling point adaptively.Secondly,manifold distance instead of euclidean distance was defined to calculate geodesic distance furtherly.So the submanifold character could be extracted from the training sample.Standardized monitoring statistics were used in process monitoring application and fault detection.Finally,the similarity index was used for pattern matching in the sub-manifold fault database.Simulation results of TE（Tennessee Eastman） process show that improved dynamic isometric mapping（IDISOMAP）-based method is more effective for fault detection and fault identification.