基于改进BP神经网络的模拟电路故障诊断

Fault diagnosis of analog circuits based on improved BP network

针对BP神经网络存在的固有缺陷:收敛速度慢,容易陷入局部极小,文中对所建BP网络的学习算法进行了改进,采用附加动量项和自适应调整学习率的BP算法对网络进行训练,替代标准BP算法中的梯度下降法寻找最优网络连接权值。仿真实验证明,这种学习算法提高了BP网络的学习效率及稳定性,大大提高了网络的收敛速度,更好地实现了对模拟电路的故障诊断。

For the inherent shortcomings of BP neural network：slow convergence and easy to fall into local minimum,in this paper the learning algorithm of established network is improved.Replacing the gradient descent method of the standard BP algorithm,the network is trained by additional momentum term and adaptive learning rate BP algorithm to find the optimal network connection weights.Simulation results show that this learning algorithm improves the learning efficiency and stability of BP network and accelerates the convergence speed of the network,and better realizes the fault diagnosis of analog circuits.