基于启发式学习算法的空间柔性机械臂神经网络控制

针对非线性强耦合的空间漂浮基柔性机器人,提出了一种启发式学习算法的神经网络的前馈控制策略。首先通过拉格朗日法和假设模态方法建立了漂浮基柔性空间机器人的动力学模型,然后采用两个神经网络及一个PID控制器来构建前馈在线学习控制系统,其中一个神经网络充当前馈控制器,另一个神经网络通过学习逆动态模型来为前馈控制器提供在线学习参数,而PID控制器主要作为辅助补偿控制器。该控制策略不是在PID控制器的指导下进行学习,且无需预先的离线学习,因而学习精度更高,且减少了对学习样本选择不当的影响,采用固定中心参数,而扩展宽度采用启发式关系确定,网络权值采用改进的最优准则算法进行调整的算法来实现快速学习能力,具有较好的实时性。仿真结果证明了所提方案的有效性。

Heuristic Learning Algorithm of Neural Network Control for Space Flexible Manipulators

According to nonlinear strong coupling of free-floating space robot with flexible manipulators,a neural network feedforward control scheme based on heuristic learning algorithm is proposed.Firstly,the dynamics of free-floating space flexible robot is established by the Lagrange and assumed modes methods.And then two neural network controllers and one PID controller are used to construct feedforward on-line learing control system.One neural network controller is used as a feedforward controller,and the other is applied to study inverse-model of the robot and provide on-line learning parameters for the feedforward controller.The PID controller is an assistant controller.The propsed control strategy can improve the control accuracy and the asymptotic convergence of tracking error,without the help of PID and off-line learning.The neural network parameters can adjust adaptively by adopting the heuristic learning algorithm.The learning speed of the neural network is enhanced greatly and has better real time property.Finally,The simulation results have shown that the presented strategy is effective.