摘要
利用凸集投影理论深入研究了函数链神经网络（ＦＬＮＮ）的学习问题，给出了相应的学习算法，并对算法的松弛形式进行了详尽的分析。由于采用了投影技术，网络的学习速度要比δ规则快得多，本文最后给出的仿真结果验证了该算法的稳定性和有效性。
In this paper, the problem of learning about function link neural network(FLNN) is studied deeply using the principle of projection on convex sets(POCS), and the corresponding training algorithm is proposed At the same time, the relaxation form of the algorithm is further analysed in detail The FLNN is trained faster based on the algorithm using projection technology than using δ rule Finally, the simulation results have verified the stability and validity of the algorithm
出处
《中国铁道科学》
EI
CAS
CSCD
北大核心
1998年第3期10-17,共8页
China Railway Science
基金
国家博士后科学基金
