The dissolved gas analysis in transformer oil has been used widely in substation. However

some online monitoring devices often have abnormal or missing data

which affects the real-time and accurate judgment of status of transformer. Therefore

it is urgent to repair the removed ‘dirty data’ and missing data in the on-line monitoring data. On the basis of summarizing the characteristics of on-line monitoring data of the dissolved gas of transformer oil at site and considering the requirements of timeliness and accuracy of the data repair

a repair strategy consisting of such three algorithms as sliding average

radial basis function neural network and polynomial fitting is proposed in this paper. The repair effect

optimal parameters

advantages and disadvantages of the three methods and their mutual cooperation are analyzed by the use of the typical filed data

and the fast and accurate repair of the on-line monitoring data of the dissolved gas in oil is achieved.