基于特定随机输入样本的模糊δ规则的收敛性

摘要

Convergence for fuzzy S rules in a special stochastic input sample is presented under the fuzzily separable condition. Here special stochastic order means that the samples supplied to the network is cycle by cycle, and in each cycle samples is stochastically chosen. Weights sequence is proved to be monotonous decrease in the training process and learning rates are provided to guarantee its convergence.%讨论在样本模糊可分条件下,基于特定随机输入样本的模糊δ规则的收敛性.基于特定随机输入是指样本按轮次输入网络,每一轮按照随机排序选取样本.证明在训练过程中权值单调下降,并最终达到收敛,给出了学习步长的选取范围.