基于深度神经网络模型的无铁心永磁同步直线电机结构优化研究

Structural Optimization of Air-core Permanent Magnet Synchronous Linear Motors Based on Deep Neural Network Models

针对无铁心永磁同步直线电机(permanent magnet synchronous linear motor,PMSLM)存在推力波动问题,该文从电机结构优化着手,引入深度学习算法建立PMSLM快速计算模型,并进行全局优化以实现推力波动抑制。首先,通过有限元模型获取PMSLM结构参数与推力及推力波动的样本数据,采用深度神经网络(deep neural network,DNN)建立其非参数快速计算模型,并与K近邻域算法和支持向量机建模方法对比,验证DNN优越性;其次,以"推力密度不削弱,推力波动最小"为目标,采用免疫克隆算法对电机结构参数进行多工况优化;最后,仿真分析和样机测试验证该方法的正确性和有效性。

Aiming at the problem of thrust ripples in air-core permanent magnet synchronous linear motors(PMSLM), this paper started with the optimization of motor structures, introduced deep learning algorithms to establish the PMSLM fast calculation model, and carried out global optimization to realize the suppression of thrust ripples. Firstly, through the finite element model, a sample data of the relationship between PMSLM structure parameters and thrust and thrust ripples was obtained, and the non-parametric fast calculation model was established by the deep neural network(DNN), which was compared with the K-nearest neighborhood algorithm and the support vector machine modeling method to verified the superiority of DNN. Secondly, aiming at "no reduction of thrust density and minimum thrust ripples", the immune clone algorithm was used to optimize the structure parameters of the motor under multiple operating conditions. Finally, simulation analysis and prototype tests verified the correctness and effectiveness of the method.