摘要
利用傅里叶级数的原理,构造单输入、多输出(SIMO)傅里叶神经网络,将非线性映射转化成为线性映射,将求解神经网络权值的方法由非线性优化方法转化成为线性优化方法,并采用最小二乘法计算网络的权值,从而大大提高了神经网络的收敛速度并避免了局部极小问题.而且,在训练输出样本受白噪声影响时,最小二乘法具有良好的降低噪声影响的功能.
Based on Fourier series principle, the single input, multiple outputs (SIMO) Fourier neural networks are proposed. The SIMO Fourier neural networks turn nonlinear mapping relationship into linear mapping relationship , turn the method of solving neural networks' weights from the nonlinear optimization method to linear optimization method, and use the least square method to compute the weights of the network. So, the SIMO Fourier neural networks highly improve the convergence speed and avoid local minima problem. When the training output samples are affected by white noise, the least square method have good denoising function.
出处
《信息与控制》
CSCD
北大核心
2004年第3期347-351,共5页
Information and Control
关键词
傅里叶神经网络
非线性优化
线性优化
最小二乘法
Fourier neural network
nonlinear optimization
linear optimization
least square method
