【摘要】 提出了一种D-FNN的新算法。其算法的最主要特点是:D-FNN选择高斯函数作为网络的激活函数和模糊系统的隶属函数,该算法不仅具有强大的全局映射泛化能力,而且在细化局部方面也有效;使用特征值分解修剪技术使得网络结构不会持续增长,可获得更为紧凑的D-FNN结构,避免了过拟合现象。最后通过对Her-mite多项式逼近能力来验证所提方案的有效性。仿真结果表明使用特征值分解修剪技术和高斯激活函数的D-FNN具有良好的性能。更多还原

【Abstract】 A new algorithm,which uses D-FNN Gauss function as a network activation function and fuzzy membership function,is proposed.The algorithm obtains strong global mapping generalization ability and effectiveness in local refinement,and adopts eigenvalue decomposition pruning technology to enables more compact D-FNN structure to avoid the phenomenon of over fitting.Finally,the algorithm is confirmed through the Hermite polynomial approximation to approach ability of validity.The simulation results show that the eigenvalue decomposition of pruning techniques and Gauss activation function of D-FNN has good performance.更多还原