[1]何新贵,许少华.过程神经元网络及其在时变信息处理中的应用[J].智能系统学报,2006,1(01):1-8.
 HE Xin-gui,XU Shao-hua.Process neural networks and its applications in time-varying information processing[J].CAAI Transactions on Intelligent Systems,2006,1(01):1-8.
点击复制

过程神经元网络及其在时变信息处理中的应用(/HTML)
分享到:

《智能系统学报》[ISSN:1673-4785/CN:23-1538/TP]

卷:
第1卷
期数:
2006年01期
页码:
1-8
栏目:
学术论文—智能系统
出版日期:
2006-03-25

文章信息/Info

Title:
Process neural networks and its applications in time-varying information processing
文章编号:
1673-4785(2006)01-0001-08
作者:
何新贵许少华
北京大学视觉听觉智能信息处理国家实验室,北京100871
Author(s):
HE Xin-guiXU Shao-hua
National Laboratory on Machine Perception,Peking University, Beijing 100871,China
关键词:
过程神经元网络时变系统信息处理学习算法仿真试验
Keywords:
process neural networks time-varying system information processing learning algorithm simulation experiment
分类号:
TP18
文献标志码:
A
摘要:
针对时变信息处理和动态系统建模等类问题,建立了输入输出均为时变函数的过程神经元网络和有理式过程神经元网络2种网络模型.在输入输出为时变函数的过程神经元网络中,过程神经元的时间累积算子取为对时间的积分或其他代数运算,它的时空聚合机制和激励能同时反映外部时变输入信号对输出结果的空间聚合作用和时间累积效应,可实现非线性系统输入、输出之间的复杂映射关系.在有理式过程神经元网络中,其基本信息处理单元为由2个成对偶出现的过程神经元组成,逻辑上分为分子和分母2部分,通过有理式整合后输出,可有效提高过程神经元网络对带有奇异值过程函数的柔韧逼近性和在奇异值点附近反应的灵敏性.分析了2种过程神经元网络模型的性质,给出了具体学习算法,并以油田开发过程模拟和旋转机械故障诊断问题为例,验证了这2种网络模型在时变信息处理中的有效性.
Abstract:
Aimed at the problems of the timevarying information processing and the dynamic system modeling, two kinds of process neural network models, including the rational formula process neural networks and the process neural ne tworks with time-varying inputs and outputs function,were built in this paper. In the process neural networks with timevarying inputs and outputs function, the time accumulation operator of process neuron was adopted as the integral to time or other algebra operations, its spacetime aggregation mechanism and incitation could synchronously reflect the space aggregation and stage time accumulation effect of exterior timevarying input signals to the output results, so as to com plete the complex mapping relationship between the inputs and outputs of non-linear system. In the rational formula process neural networks, its basic information processing unit was made up of two process neurons which appear dually,and logi cally divided into numerator and denominator, then output after rational formula combining, it can effectively advance the flexile approximation of process neural networks to the process functions which have singular values and the facility of reaction nearby the singular value point. The characteristics of these two kinds of process neural networks models were analyzed in this paper, the concrete learning algorithms were given, the effectiveness of these two kinds of network modes in time-varying information processing was proved by the cases of the process simulation and fault diagnosis of rotating machinery in the oil filed exploitation.

参考文献/References:

[1]高    隽. 智能信息处理方法导论[M].北京:机械工业出版社,2004.
GAO Jun. The introduction of intelligence information processing technology[M ]. Beijing: China Machine Press,2004.
[2]MCCULL OCH W S,PITTS W H.A logical calculus of the ideas immanent in neuron activity[J]. Bulletin Mathematical Biophysics, 1943, 5(1): 115-133.
[3]WAIBEL A. Phoneme recognition using time delay NN[J]. IEEE Trans ASSP, 1989,37(2):328-339.
[4]TSOI A C. Locally recurrent globally feedforword networks[J].IEEE Transac tions on Natural Networks, 1994 (5):229-239.
[5]DRAYE J S. Dynamic recurrent NN:A dynamical analysis[J]. IEEE Trans SMC(B),1996,26(2):692-706.
[6]HERTS J. Introduction to the Theory of Neural Computation[M].Boston :Addisonwesley,1991.
[7]欧阳楷,刘卫芳. 基于生物的神经网络的理论框架——神经元模型[J]. 北京生物医学工程,1997,16(2):93-101.
OUYANG Kai, LIU Weifang. The biologybased theoretical frame for the neural network—neuron model[J]. Beijing Biomedical Engineering, 1997,16(2):93-101.
[8]ZHANG Li, TAO Huizhong, HOLT C E ,et al. A critical window for coope r ation and competition among developing retinotectal synapses[J]. Nature, 1998, 395: 37-44.
[9]HE Xingui, LIANG Jiuzhen. Process Neural Networks[A]. Proceedi ngs of Conference on Intelligent Information Processing[C]. Beijing, 2000 .
[10]何新贵,梁久祯. 过程神经元网络的若干理论问题[J]. 中国工程科学,20 00, 2(12):40-44.
 HE Xingui, LIANG Jiuzhen. Some theory issues of process neural networks[J]. En gineering Science, 2000, 2(12):40-44.
[11]何新贵,许少华. 一类反馈过程神经元网络模型及其学习算法[J]. 自动化学报,2004, 30(6):801-806.
HE Xingui, XU Shaohua. A feedback process neuron network model and its learning algorithm[J]. Acta Automatica Sinica, 2004, 30(6):801-806.
[12]许少华,何新贵. 自组织过程神经网络及其应用研究[J]. 计算机研究与发展,2003, 40(11):1612-1615.
XU Shaohua, HE Xingui. Research and applications of selforganization process neural networks[J]. Journal of Computer and Development, 2003, 40(11): 1612-1615.
[13]许少华,何新贵. 径向基过程神经元网络及其应用研究[J]. 北京航空航天大学学报, 2004, 30(1):14-17.
XU Shaohua, HE Xingui. Research and applications of radial basis process n eur al networks[J]. Journal of Beijing University of Aeronautics and Astronautics, 2004, 30(1): 14-17.
[14]王能超. 同步并行算法设计[M]. 北京: 科学出版社, 1996.
WANG Nengchao. The design of synchronous parallel algorithm[M]. Beijing: Scien ce Press, 1996.

相似文献/References:

[1]吴琪,李晔.基于四元数的欠驱动AUV的镇定控制设计[J].智能系统学报,2014,9(02):186.[doi:10.3969/j.issn.1673-4785.201210058]
 WU Qi,LI Ye.Stabilization design of underactuated AUV based on quaternion[J].CAAI Transactions on Intelligent Systems,2014,9(01):186.[doi:10.3969/j.issn.1673-4785.201210058]

备注/Memo

备注/Memo:
收稿日期:2006-02-23.
基金项目:国家自然科学基金资助项目(60373102,60473051);教育部博士点基金资助项目(20030001701).
作者简介:
何新贵,男,1938年生,北京大学教授,博士生导师,中国工程院院士,北京计算机学会理事长.《计算机学报》副主编.主要研究方向为模糊逻辑、神经网络、进化计算、数据库理论,发表论文130余篇,其中多篇被SCI.EI检索.
许少华,男,1962年生,博士后,教授,博士生导师.研究方向为模式识别、神经网络、智能信息处理.在国内外学术期刊发表论文50余篇,其中20多篇被SCI 、EI检索.
更新日期/Last Update: 2009-04-06