基于多维离散傅立叶变换的神经网络用于数据逼近和泛化建模

A Neural Structure Based on Multidimensional Discrete Fourier Transform For Modeling of Approximation Generalization

文章在一种基于多维离散傅立叶变换原理的前馈神经网络模型的基础上，提出了一个适用于高精度逼近和泛化建模的神经网络。它可用来逼近任何连续函数，且逼近精度高，泛化能力强，学习速度快。计算机模拟实验结果显示了该网络较好地利用了离散傅立叶交换在结构和性能上的优点，在天线罩视线误差校正建模研究上很好的效果。

A new feedforword neural network structure based on the multidimentional Discrete Fourier Transform (DFT) has been used in this article.Due to its structure,the network can approximate and deduce any continuous function.The accuracy of function approximation and deducing are quite high and the learning speed is quite fast. The simulation results show that this kind of DFT neural network can take the advantage of its structure and performance,and get a good result in the modeling of error correcting.