【作者】 孙燕楠； 韩敏；

【Author】 Sun Yannan, Han Min(School of Electronic and Information Engineering. Dalian University of Technology. Liaoning Dalian 116023)

【机构】 大连理工大学电子与信息工程学院自动化系；

【摘要】 本文利用Levenberg-Marquardt算法训练三层前向神经网络,实现模糊if-then规则集。针对不同形式的规则集,提出应用区间训练法或参数训练法改变神经网络的输入输出形式。区间训练法,利用不同的隶属度对应的语言值隶属度函数的论域区间作为网络的输入输出;参数训练法,利用代表语言值隶属度函数的参数作为网络的输入输出。仿真过程中,比较了两种方法对规则的逼近能力、计算时间以及泛化能力。指出两种方法适用于不同的规则集,都具有较好的逼近和泛化能力,参数训练法比区间训练法要花费更少的时间。更多还原

【Abstract】 The paper uses a three-layer feed-forward neural network with Levenberg-Marquardt learning algorithm to realize fuzzy if-then rules. For different types of rule sets the paper presents two methods the interval-training method and the parameter-training method to change the inputs and outputs of the network. The interval-training method. intervals corresponding different memberships are as inputs and outputs of the network. The parameter-training method. the parameters representing the membership function of the linguistic value are as inputs and outputs. During simulation we compare the two methods in terms of the approximation ability to rules, the computation time and the generalization ability. Simulation results show that the two methods are suitable for different rule-sets and both have good approximation ability and generalization ability. While the parameter-training method costs less time to approximate rule-sets.更多还原