最优加权随机汇池网络的自适应算法研究  

Study of Adaptive Algorithms for Optimally Weighted Stochastic Pooling Networks

作　　者：韩博 景文腾 耿金花[1] 段法兵[1] HAN Bo;JING Wenteng;GENG Jinhua;DUAN Fabing(Institute of Complexity Science, Qingdao University, Qingdao 266071, China)

机构地区：[1]青岛大学复杂性科学研究所,山东青岛266071

出　　处：《复杂系统与复杂性科学》2018年第4期85-89,共5页Complex Systems and Complexity Science

基　　金：国家自然科学基金(61573202)

摘　　要：对最优加权随机汇池网络的自适应算法进行研究,以均方误差作为随机汇池网络输出性能评价指标,推导了最小均方(LMS)和Kalman-LMS算法的递归表达式,并应用到输入信号方差发生改变的非稳态情况中,结果表明两种自适应算法都能够迭代收敛到权的最优解。与LMS算法相比,Kalman-LMS算法不仅收敛速度快,而且权均方偏差每一步都是最优的,在网络节点的个数较少时,Kalman-LMS算法能够获得更小的均方误差,而随着网络节点的个数增加,两种自适应算法得到的均方误差趋于一致。In this paper, the adaptive algorithm of optimally weighted stochastic pool network is studied. The mean square error is used as the output performance evaluation index of the stochastic pooling network. The recursive expressions of the least mean square (LMS) algorithm and the Kalman-LMS algorithm are derived. The related results show that, for the nonstationary case of varying variances of inputs, both adaptive algorithms can converge to the optimal solution of weight vectors. However, the Kalman-LMS algorithm not only has a fast convergence speed, but also the weight mean square deviation is optimal at each step. When the number of network nodes is small, Kalman-LMS can obtain a smaller mean square error. As the number of network nodes increases, the mean square error obtained by the two adaptive algorithms tends to be consistent.

关 键 词：随机汇池网络 均方误差 最小均方 自适应算法 非稳态信号 

分 类 号：TN911.7[电子电信—通信与信息系统]