CAJ | 학술논문

为了解决径向基函数(RBF)神经网络权值与结构难以确定的问题，基于权值直接确定法，及隐层神经元中心、方差、数目与神经网络性能的关系，提出一种边增边删型的网络权值与结构双确定法。在此方法基础之上，构建一种 RBF神经网络分类器并探讨其分类性能和抗噪能力。计算机数值实验结果验证所提出的边增边删型的权值与结构双确定法能够快速、有效地确定网络的中心、方差和网络最优的权值与结构，所构造的模式分类器具有优越的分类性能和抗噪能力。
위료해결경향기함수(RBF)신경망락권치여결구난이학정적문제，기우권치직접학정법，급은층신경원중심、방차、수목여신경망락성능적관계，제출일충변증변산형적망락권치여결구쌍학정법。재차방법기출지상，구건일충 RBF신경망락분류기병탐토기분류성능화항조능력。계산궤수치실험결과험증소제출적변증변산형적권치여결구쌍학정법능구쾌속、유효지학정망락적중심、방차화망락최우적권치여결구，소구조적모식분류기구유우월적분류성능화항조능력。
In order to solve the difficulties in determining the weights and structure of the radial basis function (RBF) neural network.Based on the weights-direct-determination (WDD)method and the relationship among centers,variances, the number of hidden-layer neurons and the performance of the neural network,a pruning-while-growing-type weights-and-structure-determination (PWGT-WASD)algorithm is proposed.On the basis of the PWGT-WASD algorithm,a kind of RBF neural network classifier is constructed,and its classifying and antinoise ability is further discussed in this paper.Com-puter numerical experiment results substantiate that the proposed PWGT-WASD algorithm can determine the centers,the va-riances and the optimal weights and structure of RBF neural network quickly and effectively.The constructed RBF pattern classifier has the superiority in terms of classification and antinoise ability.