Research On Neural Network Based Intelligent Control Method For 3-Dof Helicopters

Recently,small UAV has been popular in commercial fields because of its mobility.The number of UAVs is increasing rapidly.With the brilliant commercial prospect,the control of aircraft is receiving more and more attention.An appropriate control algorithm suits the hardware and helps in getting the best performance of the system.Meantime,adaptability of the intelligent control algorithm can handle with the inaccurate part of the model,which makes it possible to obtain the most suitable control algorithm.The three-degree-of-freedom helicopter platform selected in this paper has the characteristics of typical UAVs and is friendly in data collecting.It is an ideal platform for the application of intelligent control algorithm.With this platform,this paper mainly focuses on two kinds of intelligent control methods based on neural network,adaptive backstepping method and reinforcement learning method.The primary contributions include the following aspects.Firstly,the controller is designed with the backstepping technique in which the neural networks are used to approach the nonlinear part of the system.The stability of the designed system is proved.Experiments are carried out on the simulation and physical platform of the three-degree-of-freedom helicopter.A good control performance has been achieved.Secondly,in the reinforcement learning method,the actor-critic algorithm is used.The dynamic index is introduced to improve the traditional binary reward,which solves the problem of balancing the large initial error and the small operation error in the process of selecting the threshold.The improved method is trained and verified on the simulation platform,and the controller successfully tracks time-varying signals.In all,two intelligent control methods based on neural networks are successfully applied to the three-degree-of-freedom helicopter.A good control effect is obtained.It shows the feasibility of applying intelligent control method based on neural networks to small unmanned aerial vehicle(UAV).