In the paper, a novel controller is proposed to reach better control performance for manipulator system with unknown dynamics. We notice the initial weights of neural network may have influence to control performance in neural network controller. A casual selection of the initial weights may cause poorer production quality and result in consumption of more energy. A natural way is to give neural network values close to the ideal values, which is developed through estimating system parameters by common adaptive controller. In the proposed controller, precautions are considered in case of large tracking error, neural network approximating capability is ensured, input dimension is reduced and a novel weights decay term is given. Simulation and experiment results show that these techniques result in improvement in accuracy and saving of energy.

中文翻译：

---

基于模型的权重初始化方法的机械手事件触发式自适应神经网络控制

在本文中，提出了一种新颖的控制器，以在动力学未知的情况下达到更好的控制性能。我们注意到神经网络的初始权重可能会影响神经网络控制器的控制性能。随意选择初始重量可能会导致生产质量变差并消耗更多能量。一种自然的方法是使神经网络值接近理想值，这是通过使用通用自适应控制器估算系统参数来开发的。在提出的控制器中，考虑到大的跟踪误差时的预防措施，确保了神经网络的逼近能力，减小了输入尺寸，并给出了一个新的权重衰减项。仿真和实验结果表明，这些技术可以提高精度并节省能源。