| The flying inverted pendulum system as an underdrive system,has strong coupling and non-linearity.It has eight degrees of freedom(DOF),a quadrotor with six degrees of freedom,and an inverted pendulum with two degrees of freedom.The combination of them as a new controller platform not only retains the characteristics of the inverted pendulum,but also introduces the model uncertainty of the quadrotor,which meets the verification requirements of the current control methods of complex practical systems.At the same time,it is also conducive to studying the high-precision flying attitude of the drone and improving the working efficiency of the drone.In existing research fields,designing controllers is based on linearized mathematical models.However,the actual controller form is complex and the parameters are difficult to adjust,which requires a lot of experimentation.In addition,the controller designed with inappropriate parameters is susceptible to external interference.In view of the above problems,this paper proposes a controller based on genetic algorithm or reinforcement learning algorithm.Through simulation experiments,the feasibility of the algorithm is verified.The main research work of this paper includes the following aspects:(1)First,the dynamic equations of the system are established by Lagrange mechanics.The dynamic model of the system mainly has two aspects,one is the mathematical model of the quadrotor,and the other is the mathematical model of the inverted pendulum.Linearize the established mathematical model near the equilibrium point of the system,which simplifies the mathematical model of the system.(2)Secondly,in order to solve the problem that the controller parameters are difficult to adjust,we use genetic algorithms to optimize the LQR controller.Compared to the trial and error method or the empirical method,genetic algorithm,as a type of artificial intelligence algorithm,uses group search technology.It does not require the derivative value of the objective function and some other auxiliary information.Genetic algorithms are good at finding optimal values,especially in complex problems.But its convergence speed is relatively slow,and it is easy to fall into the problem of local optimal value.So we propose search models for genetic algorithms,such as models for QR chromosomes,and so on.In this way,rapid optimization of the genetic algorithm is realized,and the effectiveness and feasibility of the improved genetic algorithm are verified in simulation experiments.(3)Then,we introduce reinforcement learning algorithms so that the system can control the system to reach a stable and balanced state without relying on accurate mathematical models.We applied the Q-learning algorithm.It can solve equations online by measuring real-time data along the system trajectory without a system model.We take the performance index in the linear quadratic programming regulator as the value function of the Q function,and obtain the Bellman equation and the Hamilton equation on the matrix.We also use the temporal difference methods to design the strategy iteration and use the iteration of the control action as an improvement step,so that the current value function and the improved control strategy can be evaluated simultaneously.Then the rewritten Q-learning is applied in simulation experiments to train the system.We have proved through a large number of simulation experiments that reinforcement learning can effectively control the stable operation of the system,which has strong robustness and anti-interference.(4)Finally,we verify whether the drone’s inverted pendulum system can run stably under the two algorithms in simulation software.By adding disturbances and noise,compare their robustness and immunity to interference.The comparison and analysis of the experimental results show that the controller based on reinforcement learning algorithm can meet the requirements of the flying pendulum system,and has stronger anti-interference ability and robustness than the genetic algorithm. |
PDF Full Text Request