In view of online monitoring data quality of dissolved gas in transformer oil

the annual data of more than 200 monitoring devices in 2020 are counted and three main data error patterns are summarized. The identification strategy and characteristic parameters of data errors pattern are proposed

and a multi-classification support vector machine is constructed to identify and classify the data errors. The Kernel principal component analysis and permutation and combination traversal method are used to reduce dimension of the feature vector. The identification accuracy of the constructed multi-classification support vector machine classifier for H

2

error data is 97.5%

and for other gasses is more than 90%. The constructed classifier is used to make statistics on the annual data of 2020

in which the error data of H

2

and C

2

H

2

reach 27.14% and 1.75% respectively.