摘要
为了提高多层前馈神经网络的权的学习效率。通过引入变尺度法,提出一种新的学习算法。理论上新算法不仅具有变尺度优化方法的一切优点,而且也能起到Kick—Out学习算法中动量项及修正项的相同作用,同时又克服了动量系数及修正项系数难以适当选择的困难。仿真试验证明了新学习算法用于非线性动态系统建模时的有效性。
A new learning algorithm is proposed by introducing the variable-schedule method in training of feedforward neural networks. In addition to the advantages the variable-schedule method has, this new learning algorithm is shown to have the same as the effects of the momentum term and the correction term used in the Kick -Out learning algorithm, and can solve the difficulties of determining the learning parame-ters in the Kick -Out algorithm. Simulation results show that the proposed method can be used effectively in the dynamic system modeling with neural networks.
出处
《控制与决策》
EI
CSCD
北大核心
1997年第3期213-216,共4页
Control and Decision
关键词
前馈神经网络
学习算法
变尺度方法
DFP法
feedforward neural networks, learning algorithm, variable-schedule method, DFP, BFGS