基于归一化最小均方算法的自适应核RBFNN  

Adaptive Kernel RBFNN Based on Normalized Least Mean Square Algorithm

作　　者：火元莲[1] 巩琪 齐永锋 安娅琦 HUO Yuanlian;GONG Qi;QI Yongfeng;AN Yaqi(College of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730070,China;College of Computer Science and Engineering,Northwest Normal University,Lanzhou 730070,China)

机构地区：[1]西北师范大学物理与电子工程学院,兰州730070 [2]西北师范大学计算机科学与工程学院,兰州730070

出　　处：《北京邮电大学学报》2022年第2期29-35,共7页Journal of Beijing University of Posts and Telecommunications

基　　金：国家自然科学基金项目(61561044)。

摘　　要：为了使自适应核径向基函数神经网络(RBFNN)有更好的收敛速度和稳态误差,提出了以归一化最小均方为学习算法对自适应核RBFNN进行优化的方法。在梯度下降算法的基础上,通过一个可变的步长因子,对归一化最小均方(NLMS)算法进行推导,并将其作为学习算法对自适应核RBFNN的权系数及偏差进行更新训练。在非线性系统辨识及模式分类中的仿真实验结果表明,使用NLMS学习算法训练自适应核RBFNN相较于其他学习算法下的自适应核RBFNN,具有更快的收敛速度及相对较小的稳态误差。To make the adaptive kernel radial basis function neural network(RBFNN) exhibit the characteristics of fast convergence and steady-state error, a method that optimizes the adaptive kernel RBFNN by using the normalized least mean square as the learning algorithm is proposed. Based on the gradient descent algorithm, we derive the normalized least mean square(NLMS) algorithm with a variable step factor, and use it as a learning algorithm to update the weights and the biases of the adaptive kernel RBFNN. The simulation results in nonlinear system identification and pattern classification show that using NLMS learning algorithm to train adaptive kernel RBFNN has faster convergence speed and relatively less steady-state error compared with other learning algorithms.

关 键 词：自适应滤波 RBF神经网络 归一化最小均方算法 非线性系统辨识 

分 类 号：TN713[电子电信—电路与系统]