Marine transformer is the key component of marine power system, transformer fault may lead to power system abnormality, and even cause ship and life safety accidents. In view of the limitations of the transformer fault diagnosis models based on basic intelligent algorithm, a transformer fault diagnosis model with strong optimization ability and high computational efficiency, which based on Quantum-behaved Particle Swarm Optimization (QPSO) and Kernel Extreme Learning Machine (KELM) was proposed. The UCI data set was used to test the models, experimental results showed that compared to SVM, ELM, KLEM, PSO-KELM and QPSO-KELM, QPSO-KELM had better classification performance and computational efficiency. Finally, the data set of real marine transformer faults were used to test QPSO-KELM, the average accuracy of testing set and training set were 92.15$\%$ and 97.65$\%$, and the standard deviations were 3.75$\%$ and 1.42$\%$, which fully proved that QPSO-KELM had excellent diagnostic accuracy and algorithm stability, and had a certain practical value for ships.