摘要
讨论了次胜者受罚的竞争学习规则 ,提出了基于正交最小二乘 ( OLS)递推算法 ,采用改进的 Givens旋转变换技术避免了大型矩阵的 QR分解运算。在满足系统测量精度条件下 ,使用反向优选算法优化 RBF网络结构。仿真结果表明 ,所得算法能有效地解决网络学习隐层单元的确定需要人介入的问题 。
This paper discusses rival penalized competitive learning and proposes the recursive orthogonal least squares algorithm. The use of modified Givens rotations avoids orthogonal decompositioin of complex matrices. In case of satisfying measure accuracy, backward selection method reduces the architecture of RBF networks. Simulation results show that algorithms can effectively calculate the number of hidden layer nodes on network learning without intervention and the method could be applied to nonlinear system ...
出处
《华东理工大学学报(自然科学版)》
CAS
CSCD
北大核心
2001年第5期503-506,共4页
Journal of East China University of Science and Technology