CAJ | 학술논문

针对连续紧凑型小波神经网络（WNN）收敛速度慢问题，提出了用Levenberg-Marquardt（LM）算法改进的小波神经网络LM-WNN．为了克服LM-WNN由于收敛速度过快易陷入局部最小点和平台的缺点，利用模拟退火（SA）算法对小波神经网络的参数进行优化，得到一组接近全局最小值的近似解，把近似解作为小波神经网络权值和阈值矩阵的初始值，以确保LM-WNN收敛于全局最小点．把SA-LM-WNN用于模拟电路故障诊断，仿真结果表明，该算法能够快速收敛于全局最小点，仿真效果较好．
침대련속긴주형소파신경망락（WNN）수렴속도만문제，제출료용Levenberg-Marquardt（LM）산법개진적소파신경망락LM-WNN．위료극복LM-WNN유우수렴속도과쾌역함입국부최소점화평태적결점，이용모의퇴화（SA）산법대소파신경망락적삼수진행우화，득도일조접근전국최소치적근사해，파근사해작위소파신경망락권치화역치구진적초시치，이학보LM-WNN수렴우전국최소점．파SA-LM-WNN용우모의전로고장진단，방진결과표명，해산법능구쾌속수렴우전국최소점，방진효과교호．
To solve the continuous and compact wavelet neural network(WNN)'s slow con- vergence speed problem, the Levenberg-Marquardt (LM) algorithm is proposed to improve the WNN, and in order to overcome the LM-WNN's shortcomings that it has too fast con- vergence speed, which make it easily trapped in local minimum points and platform, simu- lated annealing (SA)algorithm is used to optimize the parameters of the wavelet neural net- work, which results a group of approximate solution. The approximate solution is put as the initial value of weights and threshold matrix of the LM-WNN ,to ensure that the LM- WNN converge to the global minimum point. Apply the SA-LM-WNN in analog circuit fault diagnosis problem, simulation results show that the algorithm can quickly converge to the global minimum, and the simulation effect is good.