论文部分内容阅读
在模拟电路故障诊断中,由于标准的BP神经网络算法在训练样本时存在着收敛速度慢、分布不均匀、效率不高等缺点,导致电路的整体诊断性能下降。提出了一种将Levenberg-Marquardt(LM)算法与神经网络相结合的方法,对电路的脉冲信号进行多尺度分解,提取故障特征作为神经网络的输入对网络进行训练。实验仿真表明,Pspice与Matlab相结合的样本训练方法的稳定性高于传统方法,证明了该方法的实用性与可行性。
In the analog circuit fault diagnosis, the standard BP neural network algorithm has the disadvantages of slow convergence speed, uneven distribution and low efficiency in the training samples, which leads to the decrease of the overall diagnosis performance of the circuit. A method combining Levenberg-Marquardt (LM) algorithm and neural network is proposed. The pulse signal of the circuit is decomposed at multiple scales and the fault feature is extracted as the input of neural network to train the network. The experimental simulation shows that the stability of the sample training method combining Pspice and Matlab is higher than the traditional method, which proves the practicability and feasibility of the method.