Aiming at the problem of complex modulation characteristics of vibration signals of the planetary gearbox make the state identification model low accuracy,a state identification method for key components of planetary gearbox based on combinations of ensemble local mean decomposition（ ELMD） energy entropy and artificial fish swarm algorithm finding support vector machine（ AFSA-SVM） optimal kernel function coefficient is proposed. To begin,a number of narrow-band product function（ PF） from vibration signals are obtained by ELMD with after morphological average filter. Then the high dimensional feature vector set is built by calculating the energy entropy of the above PF. At last,the fault diagnosis model is developed based on AFSA-SVM algorithm,in which the extracted fault features are employed as inputs. The experimental results show that the proposed method can show the fault component of the original signal with effectively. It has the state identification accurate of the model is greatly improved.