Research On Control Technology Of Small Unmanned Coaxial Twin Rotor Helicopter Based On Model Predictive Control

Small unmanned coaxial twin-rotor helicopters are widely used in military,agriculture,scientific research and other fields because of their strong load capacity,compact structure,hovering,vertical take-off and landing,and high-speed flight.These applications require a coaxial twin-rotor helicopter with a reliable and stable autonomous and semi-autonomous flight control system.Therefore,studying the flight control of a small unmanned coaxial twin-rotor helicopter has great theoretical significance and application value.Due to the strong coupling,underdrive,open loop instability,time-varying parameters,strong nonlinearity,and high-order multivariable characteristics of the small unmanned coaxial twin-rotor helicopter system,the dynamics of the small unmanned coaxial twin-rotor helicopter Research on modeling and flight control technology has brought major challenges.As a cutting-edge technology for national defense,the flight control technology of unmanned helicopters has been facing technical blockades from Western countries.At present,most domestic research on the flight control technology of unmanned helicopters still stays at the level of traditional control methods,which can only meet the small maneuvering flight and near hovering point flight of unmanned helicopters.In this paper,a small unmanned coaxial twin rotor helicopter(code named C5)is taken as the research object.The model predictive control(MPC),which can solve the multivariable and state constraint problems in modern control theory,is taken as the basic control method,and combined with the extended state observer,which can observe the total disturbance of the system,ESO)and reinforcement learning(RL)algorithm,which are good at dealing with black box problem,are used to design a kind of attitude controller and position controller with good command tracking performance and strong anti-jamming performance for small unmanned coaxial twin rotor helicopter C5.Firstly,a simplified nonlinear mathematical model of small unmanned coaxial twin rotor helicopter C5 is established,and the model is further linearized into linear time invariant model(LTI model)and linear time-varying model(LTV model)according to Taylor formula.The time domain model verification of the nonlinear model shows that the dynamic matching degree between the model and the real helicopter is high.Then,according to the algorithm formula of linear time-varying model predictive control based on state space equation derived in this paper,three kinds of controllers are designed for the attitude loop of coaxial twin rotor helicopter C5,including the LTI MPC attitude controller using the LTI model as the predictive model,the LTV MPC attitude controller using LTV model as prediction model and the ESOLTV MPC attitude controller with ESOLTV model as the predictive model proposed in this paper considering the total disturbance of the system.Through numerical simulation and 3-DOF helicopter platform test,the command tracking performance and anti-jamming performance of the three controllers are compared.It is concluded that different predictive models will lead to different performance of predictive control algorithm.It is verified that esoltv MPC control algorithm proposed in this paper has better command tracking ability and stronger anti-jamming ability than general MPC algorithm.Finally,combining the reinforcement learning algorithm DDPG algorithm which can output continuous motion with the LTI MPC algorithm,a DDPG MPC helicopter horizontal position control algorithm which can realize real-time adaptive adjustment of weight parameters is proposed.DDPG MPC algorithm solves the problem that the controller shows great performance difference under different reference values due to fixed output weight parameters in traditional LTI MPC algorithm.Through numerical simulation,it is proved that DDPG MPC position controller has stronger robustness and adaptive ability than traditional MPC position controller under different position reference values.