Development of a Neural Network Control System for a Micro-Flexible Robotic Arm

提出了一种基于神经网络的柔性微操作机器人参考信号自组织控制系统。该控制系统由具有反馈回路的被控对象和具有前馈回路的神经网络的串联组合组成。在该系统中，神经网络的功能作为参考输入滤波器，它组织一个新的参考信号的闭环系统。反向传播可以通过两种不同的方式进行改进：通过优化和使用更强大的优化方法。这两个新的反向传播方法与改进的性能。通过使用动量和自适应学习率，可以减少训练时间。动量降低了反向传播对误差表面中小细节的敏感性。这有助于网络避免陷入浅最小值，这将阻止网络找到更低的错误解决方案。自适应学习速率的使用试图在保持学习稳定的同时保持学习步长尽可能大。学习速率响应于局部误差表面的复杂性。Levenberg-Marquardt优化使训练时间更短。Levenberg-Marquardt优化是比梯度下降法更复杂的方法。柔性微机械臂轨迹跟踪控制的实验结果表明，所提出的控制系统是有效的，精确地跟踪参考信号。

A reference signal self-organizing control system using neural networks for flexible micro-manipulators is presented. This control system consists of a series combination of both a plant with a feedback loop and a neural network with a feedforward loop. In this system, the neural network functions as the reference input filter and it organizes a new reference signal to the closed-loop system. Backpropagation can be improved in the two different ways : by heuristics, and by using more powerful methods of optimization. This two new backpropagation methods with improved performance is proposed. By using momentum and an adaptive learning rate training time can be decreased. Momentum decreases backpropagation's sensitivity to small details in the error surface. This helps the network avoid getting stuck in shallow minima which would prevent the network from finding a lower error solution. The use of an adaptive learning rate attempts to keep the learning step size as large as possible while keeping learning stable. The learning rate is made responsive to the complexity of the local error surface. Levenberg-Marquardt optimization makes training times even shorter. Levenberg-Marquardt optimization is more sophisticated method than gradient descent. Experimental results for the trajectory tracking control of flexible micro-manipulators show that proposed control system is effective in accurately following the reference signal.