计算物理 ›› 1996, Vol. 13 ›› Issue (3): 283-288.

• 论文 • 上一篇    下一篇

托卡马克中由线积分数据重建分布的神经网络方法

朱思铮, 张磊   

  1. 中国科学院等离子体物理研究所, 合肥 230031
  • 收稿日期:1995-03-17 修回日期:1995-12-26 出版日期:1996-09-25 发布日期:1996-09-25

RECONSTRUCTION OF PROFILE FROM LINE INTEGRAL DATE USING NEURAL NETWORK IN TOKAMAKS

Zhu Sizheng, Zhang Lei   

  1. Institute of Plasma Physics, Academia Sinica, Hefei 230031
  • Received:1995-03-17 Revised:1995-12-26 Online:1996-09-25 Published:1996-09-25

摘要: 用多层前馈神经网络,处理托卡马克中由物理量沿观察弦的线积分值重建其空间分布的反演问题。用BFGS拟牛顿法大大加快了标准误差反向传播算法(BP)的收敛速度

关键词: 线积分值, 分布重建, 神经网络, BFGS拟牛顿法

Abstract: Amultilayer feedforward neural network is used to reconstruct spacial distribution from line integral data along survey chords through the plasma in tokamaks. The BFGS quasi-Newton algorithm is employed to accelerate the convergence of standard error backpropagation (BP) greatly.

Key words: line integral data, reconstruction of profile, neural network, BFGS quasi-New ton algorithm

中图分类号: