When genetic algorithm is used to optimize the excitation trajectory, it often encounters problems such as trapping into local extremes, low search efficiency, and instability. Therefor, an improved harmony-genetic algorithm was proposed. In this algorithm, harmony search was used to complete the initialization of the genetic algorithm population; according to the dispersion of the population in the evolution process, the execution orders of crossover and mutation were determined; according to the fitness of the individual,the crossover and mutation probability were adjusted in a non-linear and adaptive way.So the algorithm had strong search ability and speed. In order to verify the performance of the harmony-genetic algorithm, the excitation trajectory of the SCARA manipulator was optimized. The results show that the harmony-genetic algorithm has significant advantages in solution accuracy and convergence speed, which can provide reference for the subsequent identification of dynamic parameters of the robot.