Bayesian network is currently one of the most effective theoretical models in the field of uncertain knowledge expression and inference. Before utilizing Bayesian networks for analysis and inference, it is first necessary to obtain their network models through structural and parametric learning, and structure learning is the basis for parameter learning. Aiming at the existing firefly algorithm that does not conform to biological rules as well as learning the Bayesian network structure that has low efficiency and is easy to fall into local optimization, MGM-FA (firefly algorithm based on mutual information and gender mechanism) was designed. Firstly, the Bayesian network skeleton graph was obtained by calculating the mutual information of nodes, and the MGM-FA algorithm was driven to generate the initial population based on the skeleton graph. Secondly, a personalized Bayesian network population updating strategy based on the gender mechanism was introduced to safeguard the diversity of the Bayesian network individuals. Lastly, the local optimizer and perturbation operator were introduced to enhance the algorithm's ability of optimality seeking. Simulation experiments were carried out on standard networks of different sizes respectively, and the accuracy and efficiency of the algorithm are improved compared with existing algorithms of the same type.