动态驱动神经网络辨识永磁直线同步电动机模型

Hybrid nonlinear autoregressive neural networks for permanent-magnet linear synchronous motor identification

永磁直线同步电动机(PMLSM)模型的建立对研究其稳态特性、动态特性和控制策略都是非常重要的．本文利用动态驱动神经网络对其进行建模,并在代价函数一致的基础上加入残差分析法来辨识模型的阶次,使得神经网络具有自动识别阶次的能力．为了克服神经网络结构依靠人工试凑的不足,使用基于Hession矩阵的修剪法来优化其结构．考虑到改进BP算法(学习速率自适应、动量项的方法)的一些固有缺点,使用NDEKF(基于节点的解耦扩展Kalman滤波器算法)来训练网络．实验证明,混合网络能够准确辨识出试验样机的阶次并且输出结果与实际结果十分接近;同时将NDEKF与改进BP算法进行对比,NDEKF算法具有收敛较快、泛化能力强等特点．

The modeling of permanent-magnet linear synchronous motor is crucial to the control, static and dynamic characters analysis for the system. The model of permanent-magnet linear synchronous motor is presented in this paper by using neural networks of the nonlinear autoregressive with exogenous inputs. For the same cost function, residual signal analysis is employed to identify motor＇s order automatically. Some shortages of back-propagation algorithm are considered, so NDEKF （node-decoupled extended Kalman filter） is applied to train networks. Finally, experiment results show that the hybrid neural networks of the nonlinear autoregressive with exogenous inputs can identify object＇s order precisely, and the output of networks is very close to the experimental result. In the experiment, the performance of NDEKF is often superior to that of BP, such as it requires significantly fewer presentations of training data and shorter training time than BP does, and has the better generalization ability.