Research On Model Identification Of Vtol Tail-sitter Uav Based On Machine Learning

Model identification of UAV(Unmanned Aerial Vehicle)is one of the most important parts in the realm of UAV technology.Tail-sitter VTOL(Vertical take-off and landing)UAV is a kind of aircraft with special structure,flight mode,which makes it challenging for model identification.In this thesis,a study on model identification based on machine learning is carried out for tail-sitter VTOL UAV,which includes the following four aspects:The existing traditional solutions in the field of UAV model identification are fully investigated,and the shortcomings of the traditional solutions are analyzed and summarized.According to the characteristics of tail-sitter VTOL UAV,the importance and feasibility of the study on tail-sitter VTOL UAV model identification based on machine learning are clarified.The dynamic characteristics of tail-sitter VTOL UAV are analyzed in depth,and the range of Angle of attack and Angle of sideslip during the flight of tail-sitter VTOL UAV is determined based on the experimental data.It is pointed out that how to deal with the nonlinear aerodynamic characteristics aroused by the large Angle of attack and Angle of sideslip is the key to model identification of tail-sitter VTOL UAV.The dynamics equations of tail-sitter VTOL UAV are established.With the study of recurrent neural network and the characteristics of tail-sitter VTOL UAV,a machine learning identification method based on recurrent neural network is proposed,and the model identification of tail-sitter VTOL UAV is carried out by using this framework.Compared with the least square method,this method can better express the nonlinear aerodynamic characteristics of tail-sitter and the result is better.On the basis of the further analysis of the dynamic characteristics of tail-sitter VTOL UAV,the known part of the model is distinguished from the unknown part,and the above machine learning identification method is improved.An improved machine learning identification method based on grey box model is proposed.Using the improved machine learning identification framework,the simulation analysis and experimental study of tail-sitter VTOL UAV were carried out.The results show that the improvement can further improve the identification accuracy.