摘  要：为实现磁悬浮开关磁阻电机悬浮力和旋转力之间的解耦控制，针对传统神经网络收敛速度慢，存在过拟合和易陷入局部极小值三个问题，提出了一种基于改进LM神经网络逆模型的磁悬浮开关磁阻电机解耦控制方法。该方法将LM算法引入网络，同时采用最近邻聚类法和归一化法对将样本数据进行修剪，并采用择优原则对LM模型的选取进行优化。为验证方法的有效性，采用传统方法训练网络，对比分析新方法的收敛、泛化以及解耦效果，仿真结果表明：该方法克服了传统网络的缺点，实现了磁悬浮开关磁阻电机径向两自由度悬浮力和旋转力三者之间的精确解耦。
关键词：磁悬浮电机；开关磁阻电机；神经网络逆模型；Levenberg Marquardt优化算法；解耦

Abstract: Due to the three problems, slow convergence velocity, over-fitting and easy to fall into local minim of BP algorithm, based on the improved Levenberg-Marquardt（LM）optimization algorithm in the neural network inverse model, a new method was proposed to realize the precise decoupling control of bearingless switched reluctance motor (BSRM). In this new method, the sampled data was processed by the nearest neighbor clustering algorithm and normalized method. Then the select principle was adopted to optimize the selection of LM network. To verify the effectiveness of the method, simulation comparison was carried out and the findings shown that the new method overcame the defects of BP network and the two-dimensional suspending force and torque of BSRM were decoupled.
Key words: bearingless motor; switched reluctance motors; neural network inverse; LM algorithm; decoupling