基于LM算法的神经网络系统辨识

Neural network system identification based on levenberg-marquardt algorithm

介绍了电流变传动系统 ,并采用基于Levenberg -Marquardt(LM )算法的BP神经网络对其进行系统辨识。LM算法是梯度下降法与高斯 -牛顿法的结合。就训练次数与精确度而言 ,它明显优于共轭梯度法及变学习率的BP算法 ,适用于系统辨识。仿真结果表明LM算法可大大地提高学习速度 ,缩短训练时间 ,且辨识效果很好。

Electrorheological (ER) drive system is introduced in this paper. The system is identified utilized neural network based on Levenberg-Marquardt (LM) algorithm. The LM algorithm is the combination of the steepest decent algorithm with the Gauss-Newton algorithm. Concerned with the training process and accuracy, the LM algorithm is superior to conjugate gradient algorithm and a variable learning rate back propagation (BP) algorithm. Therefore, it can be applied to system identification. Simulation results illustrate that LM algorithm speed up learning process and reduce training time greatly. The identification effect is very good.