Aiming at the problem that the lack of stability of the extreme learning machine has a bad influence on the transformer fault diagnosis

this paper introduces the regularized extreme learning machine with better generalization ability

and proposes a transformer fault diagnosis method based on regularized extreme learning machine optimized by NSGA2.The method takes the ratio of dissolved gas-in-oil content as characteristic input

and uses NSGA2 to optimize the input layer weight and hidden layer bias generated by the regularized extreme learning machine at random

avoids the over weight matrix of the output layer

thus improving the model stability and diagnostic accuracy.The simulation results show that this method has a higher fault diagnosis rate compared with the traditional regularized extreme learning machine.It is an effective method of transformer fault diagnosis.