线性神经网络在系统辨识中的应用研究

Application of linear Neural Network in System Identification

神经网络用于系统辨识最为常用的模型为BP模型,它能辨识线性和非线性系统,但BP网络存在收敛速度慢和易陷入局部极小值的缺点,影响系统辨识的速度和精度。线性神经网络结构简单,它采用Widrow-Hoff学习规则,收敛速度快,且不存在局部极小值问题,因此,对于线性系统,考虑用线性神经网络进行辨识,实验仿真结果表明,基于线性神经网络的系统辨识具有很高的辨识速度和精度。

BP model ,the most ordinary model used in neural network based identification ,can be applied to linear and nonlinear control system identification, but BP neural network has low training speed and maybe come into local minimum, these defects will decrease the precision and speed of identification. Linear neural network ,the most simple neural network, using Widrow-Hoff training rule ,has very high training speed and can't come into local minimum, so in the case of linear system, linear neural network can be taken into account, and the simulation results indicate that linear neural network based system identification do have very high speed and precision of identification.