Deep Identifier For Dynamic Modelling And Adaptive Control Of Unmanned Helicopter

Besides the features of small scale,light weight,cheap in price,and so on,the unmanned helicopter can implement many kinds of flight tasks,such as vertical take-off and landing,fixed point hovering,obstacle avoidance maneuver,coordinating veer,and so on.These outstanding advantages make unmanned helicopter become an pop-ular vehicle with broad application prospect,such as surveying,mapping monitoring,agricultural production,military reconnaissance and other fields.Unmanned helicopter is a complicated high-order,time-varying nonlinear system,which is strongly coupled with many phenomena like aerodynamic forces,shaking,and engine dynamics.It is a meaningful and challenging work to investigate the dynamic modelling and flight con-trol of unmanned helicopter.In the field of unmanned helicopter modelling,the first principle modelling and system identification are two main methods widely adopted by many researchers.How-ever,due to the complex aerodynamics and strong coupling characteristics of the un-manned helicopter,there are still many problems with these methods.With the devel-opment of the unmanned helicopter,the requirement of the flight control in complex dynamic environments has been increasing.In view of these shortcomings,focusing on unmanned helicopter,this thesis investigates the nonlinear modelling and flight con-trol based on deep learning,including the study on the high precision nonlinear dynamic model of the unmanned helicopter,the deep identifier combined with the first-principle modelling,system identification and the deep learning method,and the design of the adaptive controller based on the deep identifier.The main work of this thesis can be concluded as follows:(1)For the complex unmanned helicopter system,the thesis establishes a nonlinear dynamic model with the first-principles modelling,system identification and the deep learning method.The deterministic part of the dynamic model is established by the first-principle modelling and traditional system identification.The dynamic behavior of hidden states(e.g.airflow)and uncertainties(e.g.shaking)in system dynamics of helicopter are described by the deep neural network.The deep identifier is established by the combination of the two parts,including deep long short-term memory identifier and deep convolutional neural network identifier.The effectiveness of the proposed method is verified by various experiments with the real-world flight data from Stanford Autonomous Helicopter Project.The experiment results show that the proposed deep long short-term memory identifier has better robustness than latest deep rectified lin-ear unit model.The deep convolutional neural network identifier not only has better robustness than latest model,the modeling precision also improves 76.60%.(2)For the flight control of the unmanned helicopter,the thesis designs an adap-tive flight controller,including a deep identifier and a backstepping-based controller.The position and attitude control of the unmanned helicopter is completed through the designed controller.The stability of the controller is rigorously proved by Lyapunov theory.It reveals that the tracking errors for both position and attitude of unmanned helicopter asymptotic converge to a small neighborhood of the origin.In order to better illustrate the performance of the designed controller,the uncertainties identified from the real flight data are added during the simulation experiment.Our controller has good stability and control accuracy under complex uncertain environment.