面向双关节机械臂的参数可调RBF神经网络控制

A new parameter-adjustable radial basis function (RBF) neural network control strategy for a double-joint manipulator is proposed to solve the problems of large initial error in the error approximation process and long time to reach the steady state in the trajectory control of the double-joint manipulator. Firstly, the central parameter of the RBF neural network is modified by a gradient descent method, so that the parameter can be adjusted according to the real-time error of the manipulator, and the online optimization of the parameter can be realized. A fuzzy compensator with adjustable input boundary is proposed to reach the goal that the actual trajectory of the manipulator approaches the ideal trajectory better. By measuring the trajectory error and the derivative of the error, the output of the compensator is transferred to the torque control module after fuzzy reasoning, so that the output torque of the manipulator is closer to the ideal value. Additionally, a genetic algorithm is used to optimize the width of the RBF neural network function. Simulation results show that after the control torque of the manipulator is adjusted by using the RBF neural network control method with adjustable parameters, the accuracy of the manipulator control process is improved by 59%, and the steady time of the manipulator trajectory tracking is shortened by 69%.