基于一种生长型神经网络的倒立摆控制

Control of Inverted Pendulum Based on a Growing Neural Network

针对倒立摆这样的典型控制问题,提出了一种在结构与规模上可生长的神经网络控制方案。网络利用细胞生长结构算法的生长机制,在工作域中实现对刺激信号的自组织模式分类,并可通过新神经元的插入,实现网络规模的生长演化。在输出域中针对控制任务采用强化Hebb学习机制,实现不同的神经元以最佳方式响应不同性质的信号刺激。最后进行了倒立摆的自学习控制的仿真实验,表明在自治地与环境的交互作用中,通过神经网络自身的发育,该方案有效地控制了倒立摆系统。

A new approach for controlling the inverted pendulum by a growing neural network is presented in this issue. The network adopts a growing algorithm from reference to Growing Cell Structures in order to perform the pattern classification in work field. This growing mechanism is able to be evolved through the continuous growing of the new nerve cell. And the reinforced Hebb Synaptic Modification is used as the self - learning method to make the neurons in different fields to respond to the different stimulus in the best way. In the end of the paper, the experimental results show that the neural network scheme can interact autonomously with the environment and control the inverted pendulum effectively by the growing manners of neural system itself.