Single-rotor UAV flow field simulation using generative adversarial networks

In recent years, with the large-scale application of unmanned aerial vehicles (UAV) in agricultural plant protection, various shortcomings of have been identified; for example, the rotor flow field of the UAV will cause drift of the droplets, resulting in waste and secondary disaster. Therefore, digital simulation has become a necessity. However, due to the complexity of the rotor flow field of the rotor UAV and the operating environment of the UAV, digital simulation is associated with a large workload and great computational costs. It is urgent to explore new models using deep learning to identify the law of the rotor flow field. To address this problem, deep learning, in combination with flow field methods, is used to explore new models in this paper. A generative adversarial network (GAN) prediction model is proposed in this paper. The GAN includes a generation network and a discrimination network. In this paper, the features of the flow field are learned by the generative network to identify deep features of the flow field. The discrimination network distinguishes between true and false pictures by extracting features during training to realize adversarial training. This model can predict the flow field by identifying features of the flow-field distribution in training samples to build a predictive model. The compression effects of the computational fluid dynamics (CFD) model and the GAN model are compared in this paper. The GAN model outperforms the CFD model in predicting the flow field and compressing the data.