Aiming at the shortcomings of Grey Wolf Optimizer（GWO）, such as it’s easy to fall into local optimum and insufficient mining capacity, the paper improve the convergence accuracy and stability of GWO based on the improvement of control factors. The wind turbine gearbox vibration signal collected by the "Gearbox Reliability Collaborative（GRC）" project of the National Renewable Energy Laboratory（NREL）in the United States was used as the analysis object. After the collection of Ensemble Empirical Mode Decomposition（EEMD）, the fuzzy entropy of each eigenmode function component was calculated and the high dimensionality was constructed. Then feature vectors are used to reduce dimensionality using isometric mapping. The improved gray wolf algorithm is used to optimize the support vector machine to diagnose the gearbox fault feature set after dimensionality reduction. The results show that compared with GWO and PSO and GA, IGWO can effectively avoid falling into local optimum and improve the accuracy and stability of SVM diagnosis. It has the highest accuracy under different test samples, and the average accuracy rate can up to 93.17%.