NDEKF神经网络辨识永磁直线同步电动机

Permanent-magnet Linear Synchronous Motor Model Using NDEKF Neural Network on Hession Optimization

将带外部输入的非线性自回归模型展开成多项式形式,在此基础上用残差分析法导出真实的阶次所满足的条件。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BP算法的一些固有缺点,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,网络的输出结果与试验样机的实际输出十分接近;同时将NDEKF与BP算法进行对比,NDEKF算法具有收敛较快、泛化能力强、不易陷入局部极小等特点。

The nonlinear autoregressive with exogenous inputs model was expanded into the polynomial function, then the condition which true ranked satisfy was presented by using residual signal analysis. In order to overcome the shortages that the design of network structure was depended on one＇s own personal experience, Hession -based network pruning was used to get the optimization network structures. Some shortages of BP （ back - propagation algorithm） were considered, so NDEKF（ node- decoupled extend Kalman filter）was applied to train networks. The experiment results showed that the hybrid neural networks of the nonlinear autoregressive with exogenous inputs can identified object＇s rank precisely, and the output of networks was very close to the experimental result. In the experiments, the performance of NDEKF was often superior to that of BP, while requiring significantly fewer presentations of training data than BP and less over training time than that of BP.