Anomaly Recognition of Non-fault Switch Action Curve Based on Image Moments and NBM
Download as PDF
DOI: 10.25236/scmc.2019.004
Author(s)
Wen Hao, Xie Guangxiang, Mao Junyun
Corresponding Author
Wen Hao
Abstract
In order to alleviate the shortcomings of the existing methods in selecting the distinguishing features of the switch action curve, reduce the impact of limited training samples, and achieve accurate multi-task classification, a new method for anomaly recognition of switch non-fault action curve is proposed. Firstly, the characteristic graph of switch action curve is defined, which is divided into five parts according to action time-sequence, and the recognition feature set is constructed by calculating the zero to second moments of switch action curve. Then, five classifiers are trained for recognition based on Naive Bayesian Model (NBM). Experiments show that the method can adapt to small-scale sample training, effectively identify the different problems existing in any action time-sequence of switch, and achieve a more comprehensive switch fault early warning.
Keywords
Switch action curve; Abnormal recognition; Image moments; NBM