The invention provides a distribution transformer fault online diagnosis method based on big data flow and migration learning, comprising the following steps: (1) carding main online monitoring quantities of a distribution transformer at the present stage, and proposing main indexes for online identification of a distribution transformer fault; (2) with the online identification index of the distribution transformer fault as an input quantity of an ARIMA algorithm, proposing a fault online identification method based on the ARIMA algorithm, and building a distribution transformer fault onlineidentification model based on a big data flow; (3) solving the distribution transformer fault online identification model based on the big data flow, and screening the distribution transformer which possibly has a fault; (4) constructing a distribution transformer fault diagnosis index system; and (5) building a distribution transformer fault diagnosis model based on a migration learning algorithmTrAdaBoost, and carrying out fault diagnosis on the distribution transformer, which possibly has a fault and is screened in the step (3). According to the method in the invention, the problem that the variety of online monitoring quantities of the distribution transformer is insufficient in richness and the problem of lack of routine test data of the distribution transformer monomer are well solved.