摘要
利用基于小波变换的单隐层前馈型神经网络，模似任意复杂的非线性映射，针对电磁机构的优化问题，使用来自有限元分析的训练信息，通过选取一簇适当的紧支正交小波基函数，合成具有一定拓朴结构的小波网络，且对网络的训练过程是一个凸函数的最优化过程，从而能得到全局最优解，学习的收敛速度很快．我们将之应用于交流真空接触器直流激磁系统的优化设计中，取得了较好的效果．
A single hidden layer feed forward neural network based on wavelet transform is discussed in this paper,which can be applied to the approximating of complex nonlinear mapping.As to a optimization problem in electromagnetics,the using of wavelet network is originated from the training information of finite element analysis,through the selection of a group of suitable orthogonal wavelet basic functions to synthesize wavelet network with a certain topological structure.Since the training problem can be transformed into an convex optimization process,the global maximum can be obtained and the learning convergence speed is increased.At last,we're applied wavelet neural network to design optimization of A.C.vacuum contactor with D.C.exciting electrical circurt and obtained satisfactory effect.
出处
《河北工业大学学报》
CAS
1996年第1期60-66,共7页
Journal of Hebei University of Technology
关键词
小波变换
前馈型
神经网络
电磁场
电器
最佳化
Wavelet transform
Feedforward neural network
Optimization design in electromagnetics.
