Aimed at the problems that in the process of bayesian network structure learning, the network structure size increases in exponential with the number of nodes, in leading to the expansion of the network structure search space, and, in turn, hampering the efficiency of network structure learning algorithms, a bayesian network hybrid structure learning method, MMPC-FPSO, is introduced in combination with maximum-minimum parent-child set constraints (MMPC) and firefly particle swarm optimization (FPSO). Firstly, in view of addressing the issues of low algorithm efficiency and inaccurate network structure due to random initialization of the network structure population in the process of bayesian network structure learning using particle swarm algorithms, a population constraint method is proposed based on the improved MMPC algorithm. Secondly, in view of tackling the problems of slow speed, low accuracy, and susceptibility to local optima in traditional particle swarm-based scoring search methods, a particle optimization strategy based on the firefly algorithm is presented. Finally, in order to validate the correctness and superiority of the proposed method, the three standard networks are applied to the structure learning. The simulation results demonstrate that the gap between the obtained BIC scores and the scores of standard networks is reduced by 68.7%, 65.5%, 34.1%, respectively by the proposed algorithm, compared to the traditional particle swarm-based structure learning methods.