Research On Autonomous Flight Strategy For Unmanned Helicopter

Unmanned helicopter has a broad application prospect in civil and military fields. High precision identification method and autonomous flight strategy are its core technologies. The experiment object is a single rotor helicopter with tail rotor. Establish motion equations of all variables based on Newton’s law and moment of momentum theorem. Considering strongly coupled and nonlinear characters of unmanned helicopter, nonlinear system identification methods are studied. Then extended least squares support vector machine(LSSVM) is proposed. Inverse decoupling method is used for attitude control. Based on triple loops control mode, PID is used to realize speed and displacement control. Finally, design a flight test to verify the proposed autonomous flight strategy.Firstly, make a detailed description of the experimental object, body coordinate system and ground coordinate system. Then make aerodynamic analysis of unmanned helicopter and establish its six-degree-of-freedom motion model.Secondly, study BP neural network and LSSVM. Analyze the regularization parameter and kernel based width of LSSVM. Then add them into the extended solution space of structural parameters. Differential evolution algorithm is used to solve these two parameters. In order to compare the identification precision of BP, LSSVM and extended LSSVM, collect flight data in hovering state to take an identification experiment.Thirdly, triple loops control mode is used to establish the control system. Attitude model working as the inner loop is controlled by inverse decoupling method. The inverse model can be identified by extended LSSVM directly. Put the inverse model in front of the attitude model to get a pseudo-linear system. To take a compromise between tracking performance and robustness, PID with a cushioning is used to control the pseudo-linear system. Speed model in middle loop and displacement model in out loop are controlled directly by PID.Select training and validation sets from actual flight data after pretreatment. Train the inverse model of attitude using extended LSSVM and establish the control system described in chapter4. Take a flight experiment to verify the autonomous flight strategy proposed in this paper.