论文部分内容阅读
提出了一种高阶CMAC(HCMAC)神经网络.它是采用高阶的径向基函数作为接收域函数,为了进一步增强对输入模式的表达,还可以用接收域函数与输入模式向量构成张量积,这时产生的是高维的增强表达,同时HCMAC沿用CMAC的地址映射方法.由于高阶接收域函数的引入,使其可以获得较CMAC连续性强且有解析微分的复杂函数近似.HCMAC在不改变CMAC简单结构的基础上较RBF网络有计算量少,学习效率高等优点.文中还首次将用于参数估计的Kalman滤波学习算法引入到这种类CMAC的网络学习中,这使HCMAC有更高的学习速度.通过仿真研究表明HCMAC除拥有CMAC和RBF网络两者的优点外,还有较这两者更多的优良特性
A high order CMAC (HCMAC) neural network is proposed. In order to further enhance the expression of the input mode, the receiver domain function and the input mode vector can be used to form a tensor product. At this time, the high-dimensional enhanced expression is generated, At the same time, HCMAC follows the CMAC address mapping method. Due to the introduction of higher-order receiver domain function, it can obtain complex function approximation with stronger continuity and analytic differentiation than CMAC. HCMAC has the advantages of less computation and higher learning efficiency than RBF network without changing the simple structure of CMAC. For the first time, the Kalman filter learning algorithm used for parameter estimation is introduced into this type of CMAC network learning, which makes HCMAC have a higher learning speed. Through simulation studies show that HCMAC in addition to having the advantages of both CMAC and RBF networks, there are more than the two excellent features