Deep learning is used for airport terminal area traffic prediction study. Neural networks have complex network topology and a large number of hyperparameters, which makes the selection of hyperparameters difficult. Previously, the hyperparameters of neural network models were usually adjusted empirically, so the particle swarm algorithm (PSO) was introduced to solve the hyperparameter search problem. To address the shortcomings and deficiencies of the basic particle swarm algorithm (Basic PSO, BPSO), a chaotic particle swarm optimization (AWCPSO) algorithm based on adaptive inertia weights is proposed. Specifically, a new dynamic adaptive inertia weight is used to strike a balance between global search and local search，and then the chaos method and particle swarm algorithm are combined to solve the problem that the particle swarm algorithm is prone to fall into local optimum. The experiments show that the AWCPSO algorithm solves the premature convergence problem of the BPSO algorithm and improves the accuracy of traffic prediction when used for hyperparameter search optimization of the AC-BiLSTM model.