Fault diagnosis of hydroelectric generating set is essentially a typical machine learning problem with small sample. As an advanced machine learning methods with outstanding performance on solving the small sample size problem，support vector machine has been employed in many fields. Nevertheless，the parameter selection of support vector machine has been remained as a problematic issue. A novel variant of artificial bee colony is proposed by introducing the levy flight strategy to tackle the parameter selection of support vector machine, and the LABC-SVM is employed in the fault diagnosis of hydropower units. The numerical simulation results show that the combination of artificial bee colony with levy flight and support vector machine can be applied to multi-class diagnosis of hydroelectric generating set and efficaciously improved the accuracy of classification, thus it is valuable for engineering.