For the purpose of tackling the difficulties of the low accuracy of transformer fault diagnosis, a transformer fault identification method based on multi-strategy improved ocean predator algorithm ( MPA ) and hybrid kernel extreme learning machine ( HKELM ) has been proposed. Firstly, kernel principal component analysis ( KPCA ) is applicable to decrease the dimension of high-dimensional linear inseparable transformer fault data and it is also used to obtain feature support data. Then, the MPA is comprehensively improved through strategies such as Bernoulli chaotic mapping, improved stage transition criterion, and best candidate to strengthen the global development ability. Finally, the improved IMPA algorithm is used to optimize the parameters of HKELM and construct the transformer fault diagnosis model. Aiming to validate the validity of the model, four transformer fault diagnosis models of HKELM optimized by common algorithms are analyzed and compared. The diagnostic accuracy of IMPA-HKELM is 94. 7%, compared with the other three basic algorithms, the diagnostic accuracy is improved by 5. 4%, 8% and 10. 7% respectively. The results show that the proposed model effectively improves the classification performance of fault diagnosis and achieves higher fault diagnosis accuracy.