电气工程学报 ›› 2017, Vol. 12 ›› Issue (10): 1-8.doi: 10.11985/2017.10.001

• 理论研究 •    下一篇

粒子群优化递归神经网络的SRM磁链观测器

王雪松,许爱德,赵中林,赵显超   

  1. 大连海事大学轮机工程学院 大连 116026
  • 收稿日期:2016-12-08 出版日期:2017-10-25 发布日期:2017-10-25
  • 作者简介:王雪松 男 1991年生,硕士研究生,研究方向为电机与控制。|许爱德 女 1974年生,教授,研究方向为电机与控制。
  • 基金资助:
    国家自然科学青年基金(51407021);中央高校基本科研业务费(3132015214)

Stator Flux Observer of SRM Based on Particle Swarm Optimized Recurrent Neural Network

Wang Xuesong,Xu Aide,Zhao Zhonglin,Zhao Xianchao   

  1. Marine Engineering College Dalian Maritime University Dalian 116026 China
  • Received:2016-12-08 Online:2017-10-25 Published:2017-10-25

摘要:

在开关磁阻电机直接转矩控制系统中,为了提高磁链观测器的性能,准确地实现磁链观测,提出基于粒子群优化的递归神经网络定子磁链观测器。利用训练样本对网络进行离线训练,通过训练过程中不断调整网络的权值和阈值,形成一个泛化能力强、结构简单的网络来实现电压、电流和磁链的非线性映射。将所建磁链观测器应用到开关磁阻电机直接转矩控制系统仿真中,仿真结果表明,对比传统递归神经网络磁链观测器,该方法不仅提高了收敛速度,而且具有很高的精确度和很强的泛化能力,证明了该方法的正确性和可行性。

关键词: 开关磁阻电机, 直接转矩控制, 定子磁链观测器, 粒子群优化, 递归神经网络

Abstract:

For improving SRM stator flux observer,a particle swarm optimization (PSO) real-time recurrent neural networks (RNN) stator flux observer is presented under the SRM DTC control system.The network is trained off-line.In the off-line training process with the training data, the number and locations of the hidden units of neural network are obtained,and during the process of the learning,the neural network is built with a much simpler and tighter structure, and stronger generalization ability to form an efficient nonlinear map,and then it facilitates the elimination of the stator flux. The SRM DTC system are simulated with the new stator flux observer, and compared with the traditional RNN stator flux observer. It is proved the new stator flux observer improves the convergence speed and also has the merits of higher precision and strong generalization ability.

Key words: Switch reluctance machine, direct torque control, stator flux observer, particle swarm optimization, recurrent neural networks

中图分类号: