一种单输入/单输出的小脑模型神经网络研究

A single input-single output cerebllar model articulation controller

针对标准小脑模型关联控制器(CMAC)存在的问题,研究了一种单输入单输出(SISO)的CMAC算法.SISO CAMC采用直接地址映射,以输入v的量化值为地址,建立起输入v与权重w的关系,并通过对输入v进行归一化,增强网络的泛化能力.另外,学习采用随机样本,具有学习简单、收敛速度快、函数逼近精度高等特点,可以在单片机上实现.最后,理论证明和仿真试验,验证了该算法的有效性.

Taking into account some flaw in the conventional Cerebllar Model Articulation Controller （CMAC）, a Single Input-Single Output CMAC Algorithm （SISO CMAC） is researched. The direct weight address mapping techniques are used in the Algorithm, and the relation between the inputs and the weights is established by taking the quantification of input v as their weight address, and the generalization capability of network is enhanced by input v normalizing. The study of network uses random sampling, so the network has characteristics of simple study and quick convergence and high accuracy of function approximation, etc, and can be implemented in single chip microcomputer. In the end, the Algorithm is proved effectively by theories and simulation experiments.