基于深层卷积神经网络的电工钢片矢量磁特性模拟

Vector Magnetic Characteristics Simulation of Electrical Steel Sheet Based on Deep Convolutional Neural Network

针对传统基于BP神经网络磁滞模型收敛速度慢、建模过程需要对磁特性的表征参数进行复杂的人工提取等问题,提出了一种基于深层卷积神经网络的电工钢片矢量磁滞特性模拟的方法,该模型将磁通密度的时间序列数据和磁场强度的幅值和相位作为卷积神经网络的训练数据,利用残差模块提高卷积神经网络的收敛速度。改进模型既可以避免传统磁滞模型中繁杂的人工特征提取的过程,也可拓展应用于特征参数不易提取的非正弦激励下磁滞模型的建立。通过对比不同网络结构的磁滞模型,改进的深层卷积网络磁滞模型既能减少迭代次数,又能保证磁滞特性模拟的精细性。

In view of such problems of traditionally BP neutral network-based magnetic hysteresis model as slow convergence speed and complicated artificial extraction of characterization parameters of magnetic characteristic in the modeling process,a kind of method to simulate the vector hysteresis characteristic on the basis of deep convolutional neural networks(DCNN)is proposed.The model takes the time series data of magnetic flux density and amplitude and phase of the magnetic field strength as the training data of the convolutional neural networks(CNN).The residual error module is used to improve the convergence speed of convolution neural network.The improvement of the model can not only avoid the complicated artificial feature extraction process in the traditional magnetic hysteresis model,but also extend and apply for the setup of hysteresis models under non-sinusoidal excitation whose characteristic parameters are not easy to extract.It is founded through the comparison of magnetic hysteresis model of different network structures that the improved DCNN magnetic hysteresis model can not only reduce the number of iterations,but also assure refinement of the magnetic hysteresis behavior simulation.