权值直解的多项式神经网络及其解释能力设计

系统仿真学报 ›› 2015, Vol. 27 ›› Issue (3): 559-570.

权值直解的多项式神经网络及其解释能力设计

沈巍, 李秋实, 宋玉坤

华北电力大学经济与管理学院, 北京 102206

收稿日期:2014-02-03 修回日期:2014-05-09 出版日期:2015-03-08 发布日期:2020-08-20
作者简介:沈巍(1965-),女,辽宁朝阳,博士,副教授,研究方向为预测理论与方法、人口预测、股指预测、智能预测;李秋实(1991-),男,吉林,研究方向为股指预测、智能预测;宋玉坤(1989-),男,辽宁丹东,硕士生,研究方向为预测理论与方法、人口预测、智能预测。
基金资助:
北京市自然科学基金(9132011)

Polynomial Neural Network with Direct Solutions and Its Interpretation of Inputs

Shen Wei, Li Qiushi, Song Yukun

Department of Economics and Management of North China Electric Power University, Beijing 102206, China

Received:2014-02-03 Revised:2014-05-09 Online:2015-03-08 Published:2020-08-20

摘要/Abstract

摘要： 设计并系统研究了广义多元多项式神经网络,单隐层广义多元多项式神经网络,证明存在最优权值向量使该网络成为未知函数的最佳逼近多项式;创造性地建立了隐层节点的自然次序上限和下限,以及重要值等概念,并引入了偏导数分析,解决了神经网络不具备解释能力的弊病。设计了权值直接解法,证明该解法所得的权值向量是迭代法逼近的最优权值向量。设计了基于Matlab的图形用户界面。通过该程序,用户可通过炒股软件更新股票数据,读取特定股票、特定容量的数据,进行不同模型下指定日期的预测。

关键词: 广义多项式神经网络, 权值直接解法, 重要值, 股指预测

Abstract: The generalized multivariate polynomial neural network and single-hidden-layer generalized multivariate polynomial neural network were designed and systematically studied, and the existence of the optimal weight vector was proved which could make the network the best approximation polynomial for an unknown function; The concepts of the natural upper and lower nodes in the hidden layer were created, and an indicator “value of importance” was creatively designed and the partial derivative analysis was introduced to solve the problem that the neural network was not able to interpret the relationship between variables. The weight vector directly was solved proving it optimal. In addition, Matlab-based graphical user interface was designed. Through this program, users could update stock data via the stock software, and predicted the stock index on specified date via different models.

Key words: generalized multivariate polynomial neural network, weights-direct-determination, importance value, stock index forecasting

中图分类号:

TP391.9

沈巍, 李秋实, 宋玉坤. 权值直解的多项式神经网络及其解释能力设计[J]. 系统仿真学报, 2015, 27(3): 559-570.

Shen Wei, Li Qiushi, Song Yukun. Polynomial Neural Network with Direct Solutions and Its Interpretation of Inputs[J]. Journal of System Simulation, 2015, 27(3): 559-570.

参考文献

[1] 袁曾任. 人工神经元网络及其应用 [M]. 北京: 清华大学出版社, 1999.
[2] 刘永红. 神经网络理论的发展与前沿问题[J]. 信息与控制, 1999, 28(1): 31-46.
[3] Simon Haykin.神经网络原理 [M]. 叶世伟, 史忠植译.北京: 机械工业出版社, 2004.
[4] 魏海坤, 徐嗣鑫, 宋文忠. 神经网络的泛化理论和泛化方法[J]. 自动化学报, 2001, 27(6): 806-815.
[5] 杨旭华. 神经网络及其在控制中的应用研究 [D]. 杭州: 浙江大学, 2004.
[6] 孙富春, 李莉, 孙增圻. 非线性系统神经网络自适应控制的发展现状及展望[J]. 控制理论与应用, 2005, 22(2): 254-260.
[7] 董聪. 人工神经网络: 当前的进展与问题[J]. 科技导报, 1999(7): 26-29.
[8] 沈政. 人工神经网络研究的局限性[J]. 心理学动态, 1991(2): 5-8.
[9] 蒲小平, 陈克兴, 戴桂康. 神经网络的哲学意义[J]. 自然辩证法研究, 1993, 9(1): 18-23.
[10] Shinying Huang, RuahuanTsaih, Wanying Lin. Unsupervised neural networks approach for understanding fraudulent financial reporting[J]. Industrial Management & Data Systems (S0263-5577), 2012, 112(2): 224-244.
[11] Kelly E Fish, Richard S Segall.A visual analysis of learning rule effects and variable importance for neural networks data mining operations[J]. Kybernetes (S0368-492X), 2004, 33(7): 1127-1142.
[12] Shee Q Wong, Nik R Hassan, Ehsan Feroz.The equity premium puzzle: an artificial neural network approach[J]. Review of Accounting and Finance (S1475-7702), 2007, 6(2): 150-161.
[13] 贺昌政, 李晓峰, 俞海. BP人工神经网络模型的新改进及其应用[J]. 数学的实践与认识, 2002, 32(4): 554-561.
[14] 沈巍. 股指波动预测模型的方法研究及应用 [M]. 北京: 知识产权出版社, 2011.
[15] Rozaida Ghazali, AbirJaafar Hussain, Nazri Mohd Nawi, et al. Non-stationary and stationary prediction of financial time series using dynamic ridge polynomial neural network[J]. Neurocomputing (S0925-2312), 2009 (72): 2359-2367.
[16] M Mehrabi, M Sharifpur, J P Meyer.Application of the FCM-based neuro-fuzzy inference system and genetic algorithm-polynomial neural network approaches to modeling the thermal conductivity of alumina-water nanofluids[J]. International Communication in Heat and Mass Transfer (S0735-1933), 2012 (39): 971-977.
[17] Ivan Maric.Optimization of Self-organizing poly neural networks[J]. Expert Systems with Applications,(S0957-4174), 2013, 40(11): 4528-4538.
[18] E Gomez-Ramirez, K Najim, E Ikonen.Forecasting time series with a new architecture for polynomial artificial neural network[J]. ScienceDirect (S1568-4946), 2007, 7(4): 1209-1216.
[19] Yunong Zhang, W Li. Genenbauer orthogonal basis neural network and its weights-direct-determination method[J]. Electronics Letters (S0013-5194), 2009, 45(23): 1184-1185.
[20] 张雨浓, 杨逸文, 李巍. 神经网络权值直接确定法[M]. 广州: 中山大学出版社, 2010.
[21] 王华成, 尹美群. BP神经网络模型对公司价值评估中自由现金流量的时序预测[J]. 统计与决策, 2005(18): 15-17.
[22] 汤凌冰, 廖福源, 罗键. 模糊神经网络在股价预测中的应用[J]. 系统工程, 2004, 22(2): 107-109.