Maneuvering Flight Control Of Fixed-wing UAV Based On Neural Network

The maneuvering flight of fixed-wing UAVs makes their movement more flexible,so that they can trace moving targets quickly in complex environments,and avoid attacks.The maneuvering flight greatly improves the combat capability and survival of the UAVs.It has a wide range of demands and applications in the aerobatic show,intelligent unmanned upgrades of retired fighter aircraft and manned-unmanned formation flights.Flight control is the basis for reliable maneuvering of UAVs,and it is also the hot topic in this field.However,the traditional control framework is mainly designed for smooth and steady flight,and many deficiencies will be present when facing the problems of nonlinear,strong coupling,Euler angle singular problems,and complex maneuver generation in UAV’s large maneuvering flight.In this paper,aiming to implement unmanned combat,a fixed-wing UAV maneuvering attitude controller and maneuver command generator are designed.A simulation platform is developed to validate the UAV autonomous maneuvering flight and the designed methods are validated by simulating.The main contents are as follows:(1)A new maneuvering adaptive dynamic inverse attitude controller is designed based on the artificial neural network(ANN)and quaternion expression.The nonlinear model of the fixed-wing UAV is established.The dynamic inverse control under the wind axis coordinate and the body axis is realized,and the strict stability proof is deduced.Then,to avoid the singular problem of Euler angle expression under large maneuvering,the control scheme of quaternion dynamic inverse is proposed.A ANN is designed to compensating the modeling error of the aircraft and the disturbance.Finally,the simulation results show that the controller can trace the attitude accurately under maneuver.(2)The long-short-time memory network(LSTM)is applied to model the maneuvering of the UAV,and the data-driven maneuver command generator with antiinterference and memory functions is realized.Based on the idea that the combat maneuver can be implement by using the local sensor data and the limited global information,the combat maneuver was modeled as a state series which can be expressed as a recurrent neural network and learned from the data.The data of the key information of the maneuvering action(attitude angle,speed,GPS,etc.)is collected in the actual flight and simulation environment respectively.The LSTM network is used to realize the maneuvering posture sequence command generation.The testing shows that the network has a good memory and anti-interference ability,which can meet the performance requirements of maneuvering instructions.(3)The complete simulation system is built and the simulation verification of autonomous flight of UAVs is realized.On the basis of the flight simulation system,the simulation implementation of various maneuvers is completed by using the above methods.The simulation results show that the designed control architecture can realize the maneuvering flight of fixed-wing UAVs.This paper systematically studies the maneuvering flight control method of UAV,and proposes a new attitude controller and command generation method.All methods have been verified by simulation which shows the effectiveness of the proposed control method.This research help to understand and realize the maneuvering flight of fixedwing UAVs.