CAJ | 학술논문

为克服误差逆向传播算法的多层前馈型BP神经网络收敛速度慢、局部极小化问题，提出用遗传算法(GA)的全局搜索能力寻求最优的BP神经网络权值和阀值，以提高神经网络的收敛速度和克服局部最优。以磁流变液压悬置动态特性试验结果为数据样本，分别用未优化的BP神经网络和优化后的GA-BP神经网络对磁流变液压悬置正、逆模型进行辨识。结果表明，相对于BP神经网络，GA-BP神经网络具有更高的辨识精度、更快的收敛速度，在磁流变液压悬置数学模型辨识方面具备更优的性能。
위극복오차역향전파산법적다층전궤형BP신경망락수렴속도만、국부겁소화문제，제출용유전산법(GA)적전국수색능력심구최우적BP신경망락권치화벌치，이제고신경망락적수렴속도화극복국부최우。이자류변액압현치동태특성시험결과위수거양본，분별용미우화적BP신경망락화우화후적GA-BP신경망락대자류변액압현치정、역모형진행변식。결과표명，상대우BP신경망락，GA-BP신경망락구유경고적변식정도、경쾌적수렴속도，재자류변액압현치수학모형변식방면구비경우적성능。
Initial weights and thresholds of BP neural network are optimized by using Genetic Algorithm(GA) method to solve its slow convergence speed and local optimum. The defect of BP neural network is thus overcome by the proposed method. The direct and inverse dynamic models for a prototype of Magneto-rheological (MR) mount are identified by using traditional BP neural network and novel GA-BP neural network. The results show that the GA-BP neural network has faster convergence rate and higher precision compared with the traditional BP neural network in the identification of direct and inverse model for MR mount.