The paper constructs a deep learning fault classification model for fault diagnosis using a Sparse Restricted Boltzmann Machine(Sparse-RBM)based on the deep learning(DP)theory, in order to synthesize multi-dimensional information, determine transformer defects quickly and accurately, and solve the problem that multi-dimensional information fusion weights are difficult to determine.Combined with the multi-dimensional monitoring of large transformers, a transformer fault diagnosis method based on multi-dimensional information fusion and deep believe network is proposed.The method can utilize the massive unlabeled multi-dimensional monitoring data of the transformer as the learning sample, and only needs a small amount of tagged data for auxiliary optimization.The trained model can make an accurate fault diagnosis of the transformer body state according to the real-time online multi-dimensional monitoring data of transformers.The diagnosis test of a 220 kV main transformer in a city is carried out.The test results show that the accuracy of the method proposed in the paper is improved by 4% compared with the existing one.