径向基函数网络能够以任意精度逼近任意连续函数。目前径向基函数神经网络中普遍使用高斯基函数作为激励函数,用它来训练网络时,首先需要确定径向基函数的扩展常数SPREAD.SPREAD设定过小,函数逼近时会出现过适性,SPREAD过大,函数逼近时会出现不适性。所以创建径向基函数网络时,需根据不同的训练集和精度要求,选择多个不同的SPREAD值来反复比较网络输出值和目标值,以确定相对较好的逼近情况,这给实际使用带来一定的困难和不便,且无法保证该选择是最佳值.本文研究使用紧支撑径向基函数作为激励函数来设计神经网络,并分析其在函数逼近中的实际应用。 Radial basis function networks can approximate any continuous function with arbitrary accuracy. Nowadays, Gaussian basis function is widely used as excitation function in radial basis function neural network. When it is used to train the network, we first need to determine the expansion constant SPREAD.SPREAD of the radial basis function is too small, too adaptability will appear when the function approximates , SPREAD is too large, there will be discomfort when the function approaches. Therefore, when creating a radial basis function network, a plurality of different SPREAD values ​​need to be repeatedly compared to network output values ​​and target values ​​according to different training sets and accuracy requirements to determine a relatively good approximation, which gives actual use of the band And can not guarantee the choice is the best value.This paper studies the use of tight support radial basis function as the excitation function to design the neural network and analyzes its practical application in function approximation.