摘要
为了克服前向神经网络的固有缺陷,提出了基于采样数据建立的含单隐层神经元的模糊前向神经网络。该网络模型利用权值直接确定法得到了最优权值,网络结构可以随采样数据的多少,自主设定隐层神经元,完成了近似插值与精确插值的转换。计算机数值仿真实验表明,模糊前向神经网络具有逼近精度高、网络结构可调和实时性高的优点,并且可以实现预测和去噪。
In order to overcome the inherent drawbacks of feed-forward neural network, based on the sampling data set, fuzzy membership function is used to construct a new neural networks with single hidden layer. For this model, the best weight is received based on the method of weights-direct- determination, and the network＇s structure can be adjusted with the change of data set for designer, and completed the conversion of approximate interpolation and accurate interpolation. The results of numerical experiment with computer show that the fuzzy feed-forward neural network has many advantages, such as high approximation precision, the structure can be adjusted, and high real-time, and can forecast and denoising.
出处
《模糊系统与数学》
CSCD
北大核心
2013年第3期122-127,共6页
Fuzzy Systems and Mathematics
基金
中央高校基本科研业务费资助项目(JCB2013B07
2011B018)
华北科技学院高等教育科学研究课题(HKJYZD201213)
关键词
前向神经网络
隶属函数
权值直接确定法
插值
Feed-forward Neural Network
Membership Function
Weights-direct-determination
Interpolation
