RBF网络做函数逼近的改进研究

Research on Approximation of Function by Improved RBF Networks

阐述了利用神经网络做函数逼近研究的必要性和合理性.通过实验发现：直接利用RBF网络（direct radial basis function networks,DRBFN）做函数及其导函数逼近时的缺点,在对此结论合理分析的基础上,给出间接利用RBF网络（indirect radial basis function networks,IRBFN）做函数及其导函数的逼近方法.仿真实验表明：改进后的网络性能有了较大提高,同时,根据实验结果提出了一个猜想.

This paper expatiates the necessity and significance of using neural networks to research function approximation. The disadvantage of function and it＇s derivatives approximation based on using RBF（direct radial basis function） networks directly is found through experiments. After analyzing the results, the paper presents a new approach called RBF（indirect radial basis function） networks. The result shows that networks have better performance.