Cascaded deep neural network framework for fixed-wing UAVs maneuvering simulation model

Abstract Fixed-wing unmanned aerial vehicles (UAVs) are characterized by their high efficiency, low cost, and flexibility, indicating a broad spectrum of developmental prospects. The construction of maneuvering simulation models for fixed-wing UAVs can effectively extend their application, hence the maneuvering model serves as the foundation of simulation. The modeling method of traditional fixed-wing UAVs maneuvering simulation model requires a lot of mathematical and physical knowledge. Therefore, the universality and reusability of the maneuvering simulation models established by these methods are poor. However, intelligent modeling methods that combine deep neural networks with physical analysis can effectively construct maneuvering simulation models for fixed-wing UAVs. A cascaded deep neural network framework for fixed-wing UAVs maneuvering simulation model is proposed. This framework adopts intelligent optimization algorithms to autonomously adjust network hyperparameters to improve prediction accuracy, and enhance the reliability of UAVs maneuvering simulation models. Through comparative experiments, among the four sets of experiments using MLP/Conv/SVM/GPR networks, the cascaded network based on the MLP network reduced approximately five orders of magnitude compared to the six errors of the integrated network. The effectiveness of the cascaded deep neural network framework and the improved particle swarm optimization algorithm have been verified, providing an efficient and accurate method for establishing a simulation model of fixed-wing UAVs.