摘要
针对过程的非线性和动态特性,提出一种基于核正交流形角不相似度的监测方法.利用两个流形子空间正交向量求取内积矩阵的奇异值,构建基于核正交流形角的不相似度指标,量化评估标准集和测试集的流形子空间的统计量关系.首先,在多流形投影方法的基础上,利用非线性函数将原始过程数据投影到特征空间;其次,引入Gram-Schmidt方法正交化投影向量,形成流形子空间的基向量;再次,对两个流形子空间的内积进行特征值分解,获得核正交流形角,构建不相似度监测模型,该监测指标融合角度和距离度量,能够更好地触发故障警报;最后,通过在TE过程上的仿真实验验证了所提出算法的优越性.
For process with nonlinear and dynamic features, a kernel orthogonal manifold angle based dissimilarity is developed to quantitatively evaluate the statistical relationship between the manifold subspaces of normal benchmark and test data sets. The kernel orthogonal manifold angle based dissimilarity index is derived from the singular values of the inner-product matrix calculated by the orthogonal vectors in the two manifold subspaces. Firstly, the historical process data is mapped into feature space by using the nonlinear function based on multi-manifold. Then, the projection vectors are orthogonalized by using Gram-Schmidt method, and base vectors of the manifold subspace are constructed.Furthermore, the kernel orthogonal manifold angle with singular value decomposition（SVD） of the inner-product of two manifold subspaces, and the dissimilarity monitoring model are got. Angle and distance measures are combined into the monitoring index to trigger fault alarm with better sensitivity. The simulation experiment on the TE process demonstrates the superiority of the proposed method.
作者
卢春红
文万志
LU Chun-hong;WEN Wan-zhi(College of Computer Science and Technology, Nantong University, Nantong 226019, China)
出处
《控制与决策》
EI
CSCD
北大核心
2018年第6期1141-1146,共6页
Control and Decision
基金
国家自然科学基金项目(61602267)
南通市科技应用项目(MS12016036)
江苏省高校自然科学基金项目(17KJB530008)
关键词
非线性动态过程监测
核正交流形投影
正交向量
不相似度指标
故障检测
nonlinear dynamic process monitoring
kernel orthogonal manifold projection
orthogonal vector
dissimilarity index
fault detection
