The current era is an era in which artificial intelligence and big data complement each other. Bayesian network, as a research hotspot in the field of AI, has good performance under the condition of uncertainty and theoretical proof. In order to better deal with the learning of Bayesian networks in big data environment, this paper proposes a Bayesian network adaptive incremental learning method based on particle swarm optimization. It learns the structure and parameters of the Bayesian network from multiple training data sets in batches by determining a novel Bayesian network structure coding, defining the computing rules of the particle swarm optimization algorithm and introducing BIC scoring function for evaluating network structures. A particle in PSO represents a network structure with two attributes of position and velocity, and the network structure with the highest fitness value is the best trained structure from the data set. This paper also explores the influence of the factors of particle swarm optimization algorithm, the number of iterations of single-batch incremental learning and the maximum number of parent nodes on Bayesian network learning under different data sets, and also improves the algorithm based on these influences. The experiment shows that the proposed method is reasonable and feasible, has high performance, and can effectively learn Bayesian network.