<p>There are some simplified parameters in the traditional dynamic model of the planetary gearbox (PGB), which leads to the consistency of simulated vibration signals with the measured vibration signals being inferior. In addition, the accuracy of the simulated signals is inferior, which reduces the accuracy of the PGB fault diagnosis based on the simulated signals. To address the above problems, a method for optimization of the PGB dynamics model based on the Improved Grasshopper Optimization Algorithm (I-GOA) and global sensitivity analysis is proposed. Firstly, an I-GOA based on adaptive social force factor and pattern search is proposed to address the problems of slow convergence speed, low search accuracy, and imbalance between global and local search ability during parameter optimization by Grasshopper Optimization Algorithm (GOA). Secondly, a sensitive feature selection method is proposed based on univariate evaluation and Pearson correlation coefficient. The method reduces the number of features for fault diagnosis, lowers the problem dimension, and is applied to the measured signals to select sensitive features for PGB faults. The method establishes a sensitive feature set to planetary gear tooth wear faults. Thirdly, to select the critical optimization parameters of the dynamical model, a model parameter selection method of Kriging surrogate model combined with Sobol’ global sensitivity analysis is proposed. Finally, the sensitivity features of the measured and simulated vibration signals are calculated by characteristic expressions, and the Euclidean distance loss of the two signals is used as the objective function of model optimization. The parameters of the PGB dynamic model are updated using I-GOA, and the dynamic model of planetary gear wear failure in PGB experimentally verifies the proposed method. Finally, the simulation experiment verifies that the proposed method is feasible and effective.</p>