Existing transformer fault diagnosis models are usually based on the correction method of numerical error

but ignored the importance of fuzzy mapping rule between equipment states and oil chromatography data. To solve the problems

we explored and improved a network structure

and proposed a self-determination and active correction of power transformer oil chromatography diagnosis model based on the deep belief network（DBN）

which can realize the automatic decision-making and switching function between the correction process based on the numerical error and the correction process based on fault difference. Firstly

a database of the faults difference degree based on space Euclidean distance was established

by which the original oil chromatography data were mapped to a high-dimensional space by a sparse auto encoder（SAE）. Then

the error correction decision unit and the multistage controllable error correction strategy based on this unit were proposed successively

which has effectively controlled the correction direction and degree of multi-layer restricted Boltzmann machine（mRBM）

realizing the self-decision between the numerical error correction process and the fault difference correction process

and improving the model’s ability to express the fuzzy mapping rule. Examples in field are used to verify the method. The results show that the model proposed in this paper can be adopted to accurately diagnose transformer faults with average accuracy of 92.32%

which can provide a reference for the operation and maintenance of substation effectively.