摘要
为了确定前向神经网络的网络结构,提出了一种基于采样数据的含单隐层神经元的模糊前向神经网络,反映了构造数据所蕴含的系统信息,其隐层神经元激励函数选择为三角型隶属函数和构造数据相应输出的乘积。该网络模型可以随采样数据的多少自主选择构造数据,自主设定隐层神经元,利用权值直接确定法得到网络最优权值。数值仿真实验表明,相比于现有文献的已有网络模型,模糊前向神经网络具有逼近精度高、网络结构可调、较好的预测性和实时性高的优点。
In order to determine the feed-forward neural network＇s structure, fuzzy feed-forward neural network was constructed based on the sampling data, which reflected the system＇s information contained in the construction data. And the hidden layer neuron activation function is the product of triangular membership function and corresponding data output. For this model, the network＇s structure can be ad- justed with the change of sampling data for designer, and the best weight was received based on weights-direct-determination. Numerical simulation results show that the fuzzy feed-forward neural network has many advantages such as high approximation precision, and the structure can be adjusted with good pre-diction and high real-time. It is better than the other feed-forward neural networks.
出处
《中山大学学报（自然科学版）》
CAS
CSCD
北大核心
2013年第2期33-37,42,共6页
Acta Scientiarum Naturalium Universitatis Sunyatseni
基金
中央高校基本科研业务费资助项目(2011B018
JCB2013B07)
华北科技学院高等教育科学研究资助课题(HKJYzd201213)
关键词
模糊前向神经网络
权值直接确定法
三角型隶属函数
实时
fuzzy feed-forward neural network
weights-direct-determination
triangular membershipfunction
real-time
