Abstract: As the empirical mode decomposition (EMD) has mode aliasing and tends to present end effects, it is difficult to effectively couple the vibration signal and extract nonlinear feature. In this work, a complete ensemble empirical mode decomposition with adaptive noise (ANCEEMD) was adopted as the signal processing method, and sample entropy was introduced for feature extraction. A neural network model optimized by swarm intelligent fusion algorithm was used to identify and diagnose the planetary gearbox faults. For shuffled frog leaping algorithm (SFLA) and particle swarm optimization (PSO) swarm intelligence algorithm, a fusion mechanism of "two-layer optimization and internal and external circulation" was implemented, and the SFLA-PSO fusion algorithm was proposed. The simulated fault tests of planetary gear were carried out with collecting the signals of multiple faults and taking out the sample entropy features. Moreover, the BP neural network model was optimized by using SFLA-PSO fusion algorithm to identify and diagnose the fault of planetary gear box. The results show that the accuracy of the SFLA-PSO-BP diagnosis model based on ANCEEMD sample entropy feature extraction is increased by 5% and 10% compared with that of PSO-BP and BP, respectively.