CAJ | 학술논문

针对传统神经网络在开关磁阻电机建模过程中存在的网络结构确定困难和训练过程过于复杂的问题，提出了基于回声状态网络的电机建模方法。回声状态网络利用储备池和线性回归算法简化了网络设计和训练过程，使得模型具有良好的收敛速度。无需电机的任何先验知识，利用训练样本，便可建立正确反映电机磁特性的电感模型。在保证良好预测精度的前提下，与BP和RBF神经网络相比，所建模型具有计算简单，收敛速度快等优势，可进一步应用于电机的实时控制中。
침대전통신경망락재개관자조전궤건모과정중존재적망락결구학정곤난화훈련과정과우복잡적문제，제출료기우회성상태망락적전궤건모방법。회성상태망락이용저비지화선성회귀산법간화료망락설계화훈련과정，사득모형구유량호적수렴속도。무수전궤적임하선험지식，이용훈련양본，편가건립정학반영전궤자특성적전감모형。재보증량호예측정도적전제하，여BP화RBF신경망락상비，소건모형구유계산간단，수렴속도쾌등우세，가진일보응용우전궤적실시공제중。
Focusing on the problems that the network structure is difficult to determining and the training process is too complex in the modeling process of switched reluctance motor using the traditional neural network, a motor modeling method based on echo state network was proposed. The echo state network simplified the network design and the training process using the reserve pond and the linear regression algorithm, enables the model to have the good convergence rate. Without any prior knowledge of the motor, inductance model would be established accurately to reflect the motor magnetic characteristics using training data. Under the premise to ensure good prediction accuracy, compared with BP and RBF neural network, the model had the advantages of simple calculation, fast convergence speed, and could be further applied to the real-time control of the motor.