For the unmanned aerial vehicle (UAV) Millimeter-Wave (mmWave) communication systems, an efficient and accurate beam training method is urgently required to overcome the severe path loss. By taking into account the mmWave propagation environment,a three-dimensional (3D) intelligent beam training strategy by leveraging the polynomial regression model and optimized beam patterns is proposed in this paper. We treat the mmWave beam selection as a polynomial regression problem. The regression function is obtained by a machine learning (ML) method based on the dataset and a special beam pattern is achieved to obtain the dataset consisting of measured powers and estimated angles. Furthermore, a noise suppression method involving the use of denoising autoencoder (DAE) is developed to improve the robustness of the proposed regression model.Numerical simulation results demonstrate that our proposed beam training strategy is capable of getting the same precision as the exhaustive search methods with a shorter time.