To ensure the fault diagnosis accuracy of an oil immersed transformer and improve convergence speed and generalizability, this paper proposes a transformer fault diagnosis method based on deep belief network and extreme learning machine optimized by a sparrow search algorithm (DBN-SSAELM). First, deep belief networks (DBN) are used to extract the features of dissolved gas data in oil. Second, an extreme learning machine (ELM) with strong learning ability is used to replace the Softmax classifier in the traditional DBN classification model to deeply analyze the correlation between features and fault types and improve convergence speed. Then, a sparrow search algorithm (SSA) is used to optimize the input weights and bias of the hidden layer node of the ELM to improve the accuracy rate and stability of the mode. Finally, the rate of accuracy, recall, precision and convergence speed of fault diagnosis are selected to evaluate the performance of the model before and after optimization. The results of transformer fault diagnosis show that the proposed DBN-SSAELM transformer model has higher fault diagnosis accuracy, stronger generalizability and better stability, and the average accuracy is 96.50%. This is suitable for transformer fault diagnosis.

This work is supported by the National Natural Science Foundation of China (No. 51807063).