The reasons for the fast loss of population diversity in bare-bones particle swarm optimization(BPSO) are analyzed, and a hierarchical learning BPSO(HLBPSO) is developed to avoid the premature convergence.In the HLBPSO, each particle can learn from the personal best(pbest) particle, the superior particle and the global best(gbest) particle according to the learning probability.The learning mechanisms can provide hierarchical searching for maintaining the swarm diversity.Moreover, the gbest particle adopts the jump strategy to strengthen the escape ability and the disturbance strategy to improve the quality of the solutions.The experimental results show that the proposed method significantly outperforms the state-of-the-art PSO algorithms in terms of convergence speed and solution accuracy.