Aimed at the problems that efficiency is low in optimization and there are defects easy to fall into local optimum in Bayesian network structure learning, a new Bayesian network structure learning algorithm based on hybrid improved Bird swarm algorithm is proposed. Firstly, the initial network is constrained by Mutual information. Secondly, the Bird swarm algorithm is improved by adding adaptive inertia weight. With the increase of the number of iterations, the adaptive inertia weight is adjusted to dynamically adjust the search space of algorithm and change the convergence speed. Finally, taking the improved Bird swarm algorithm as a search strategy, optimization is given to the structure of Bayesian network. The experimental results show that the proposed algorithm is not only good in accuracy and fast at convergence speed, but also good in global optimization ability.