Abstract: A novel object data correction based diversity neural network ensemble method is proposed by analyzing the object data correction principle. In this method, the individual networks are trained by correcting expectative output of individual network dynamically and set as new training sets, and all the individual networks in the ensemble are guided to realize diversity learning. The method is applied to the fault diagnosis of power transformer, the experiment results show that accuracy rate of the method is superior to the neural network ensemble method which is trained by individual networks, in comparison with ADL (active diverse learning), the communication cost of ensemble network is greatly reduced, it is a extremely efficient neural network ensemble method.