基于余弦相似度的复杂网络故障检测方法及应用

提出一种基于余弦相似度的复杂网络故障检测方法。利用余弦相似度确定变量之间的相关性，得到邻接矩阵，进而构建变量之间的网络模型；结合系统的网络拓扑结构，计算相应的复杂网络度量指标，对比故障状态与无故障状态下的网络结构与度量指标的差异，确定故障源；最后利用Tennessee-Eastman（TE）过程故障检测实例，结果表明，与偏相关系数方法对比，本文所提方法能有效且更准确地检测出故障。

A complex network fault detection method based on cosine similarity and its application

This paper proposes a complex network fault detection method based on cosine similarity. The method determines the correlation between variables using cosine similarity, and then generates the adjacency matrix in order to build a network model between variables, and finally calculates the corresponding metrics of the complex network according to the system network topology, thus determining the fault source for different network structures and metrics involving fault states and the fault-free state. Finally, using the Tennessee-Eastman (TE) process as an example, comparison with the partial correlation coefficient method shows that our new method can detect faults efficiently and more accurately.