摘要： 从映射的角度分析了CMAC模型各层神经元之间的关系，根据网络输入向量的量化级数、泛化参数、相邻量化级引起的重叠神经元的个数，从理论上给出了虚拟层神经元数目的范围。对于存在实际层神经元的CMAC模型，讨论了压缩映射对网络学习收敛性的影响。最后通过机器手逆运动学问题的仿真实验，进一步比较说明了在从虚拟层神经元到实际层神经元的压缩映射中不同压缩比对网络学习收敛性及系统运行精度的影响。

关键词: CMAC, 收敛性, 压缩映射

Abstract: In this paper, the relationship between every neuron layer is analyzed in the point view of mapping. According to the quantization level of input vector, generalization and the number of overlapping neurons brought by neighboring quantization levels, the maximum number of virtual neurons is given. Also, the influence on learning convergence from compression mapping in those CMAC models with actual neurons is discussed. At last, the simulation experiment of inverse kinematics of robust arm shows that different compression ratio of virtual neurons and actual ones brings different influence on CMACs learning convergence and running precision.

中图分类号: 

• TP183