摘要
采用由AND和OR模糊神经元组成的神经网络进行模糊逻辑建模,每个神经元由S和T算子组合而成,并给出单个神经元作用在模糊集上的效果,充分展示了这两种神经元的优越性,以4条规则为例推导出这类神经网络与“if-then”的规则集之间的等价关系。在神经网络的学习过程中,提出了一种混合式的学习方案采用遗传算法优化整个网络的结构,缩小了输入空间的维数,减少了相应的规则数;并在此基础上利用梯度的学习方法继续对相应的参数进行优化,从而使网络具有很好的优越性,为进一步模糊控制创造了良好的平台。
The AND neuron and OR neuron are adopted to make up of a newly neural network for fuzzy logic model in this paper,every neuron is constructed of S operator and T operator,the result of a single neuron acting on fuzzy sets is exhibited. The structure of the network directly is equivalent to a collection of ＂if-then＂ statements. The hybrid learning optimization procedure is consisted of two main phases. Firstly the structure of the whole network is optimized by genetic algorithm, the dimension of input space is reduced,and the number of rules is shortened;the second phase is aiming at refining the candidate parameters from genetic algorithm by gradient-based learning. The superiority of network is obvious.
出处
《广西师范大学学报（自然科学版）》
CAS
北大核心
2006年第4期111-114,共4页
Journal of Guangxi Normal University:Natural Science Edition
基金
交通部交通应用基础研究项目基金资助课题(200332922505)
高等学校博士学科点专项科研基金资助课题(20030151005)
关键词
模糊逻辑建模
神经网络
遗传优化
梯度优化
连通度
fuzzy logic-based models
neural network
genetic optimization
gradient-based learning
con-nectivity
