基于多层前馈神经网络SPDS算法的地质数据非线性分析问题研究

J4 ›› 2014, Vol. 36 ›› Issue (08): 1528-1532.

基于多层前馈神经网络SPDS算法的地质数据非线性分析问题研究

戴珂，张少仲，蒋波，白英，王小妹

（大连海事大学信息学院，辽宁大连 116026）

收稿日期:2012-10-22 修回日期:2013-02-25 出版日期:2014-08-25 发布日期:2014-08-25
基金资助:
国家自然科学基金资助项目（61173034）

Research of the geological data nonlinear analysis problem based
on the SPDS algorithm in the multilayer feed-forward network

DAI Ke,ZHANG Shaozhong,JIANG Bo,BAI Ying,WANG Xiaomei

(School of Information,Dalian Maritime University,Dalian 116026,China)

Received:2012-10-22 Revised:2013-02-25 Online:2014-08-25 Published:2014-08-25

摘要/Abstract

摘要：

多层前馈神经网络善于解决非线性分析问题，但对于复杂的地质数据，普通的训练算法难以收敛。首先介绍了SPDS算法，并把它用于解决地质数据的非线性分析问题。算法的仿真实验表明，用SPDS算法训练的多层前馈神经网络，比较好地解决了该问题。

关键词: BP算法, SPDS算法, 非线性分析

Abstract:

Abstract:The multilayer feedforward neural network is good at solving the nonlinear analysis problem, but for complex geological data, the common training algorithm is difficult to converge. The SPDS algorithm is proposed to solve the problem of the geological data nonlinear analysis. Simulation results show that the proposed algorithm can be better to solve the problem in the multilayer feedforward neural network.

Key words: BP algorithm;SPDS algorithm;nonlinear analysis

戴珂，张少仲，蒋波，白英，王小妹. 基于多层前馈神经网络SPDS算法的地质数据非线性分析问题研究[J]. J4, 2014, 36(08): 1528-1532.

DAI Ke,ZHANG Shaozhong,JIANG Bo,BAI Ying,WANG Xiaomei. Research of the geological data nonlinear analysis problem based
on the SPDS algorithm in the multilayer feed-forward network [J]. J4, 2014, 36(08): 1528-1532.

参考文献［9］

［1］	Zhang Ling, Zhang Bo. The analysis of BP algorithm in neural network［J］. Pattern Recognition and Artificial Intelligence, 1994, 7(3):191195. (in Chinese)
［2］	Wang Xuefeng, Feng Yingjun. A new training algorithm in multilayer neural network ［J］. Journal of Harbin Institute of Technology, 1997, 29(2):2325. (in Chinese)
［3］	Wang X F, Feng Y J, Zhao X. A single parameter dynamic searching algorithm for multilayer neural networks (Part I)［J］. International Journal of Innovative Computing, Information and Control, 2008, 4 (9):22212233.
［4］	Wang X F, Zhao X, Feng Y J. A single parameter dynamic searching algorithm for multilayer neural networks (Part II)［J］. International Journal of Innovative Computing, Information and Control, 2008, 4 (11):29412954.
［5］	Feng Yingjun, Zhang Shaozhong, Wang Xuefeng, et al. Cause for paralysis of BP network［J］. Journal of Harbin Institute of Technology, 2000, 32(5):8183. (in Chinese)
［6］	Zhang Shaozhong, Feng Yingjun, Wang Xuefeng, et al. A new algorithm of multilayer feedforward neural networks based on single parameter coordinate rotatory method［J］. Acta Electronica Sinica, 1999, 27(2):115116. (in Chinese)
［7］	Zhang Shaozhong, Feng Yingjun. Generalization and fault tolerance of single parameter dynamic search algorithm［J］. Journal of Harbin Institute of Technology, 2000, 32(6):14. (in Chinese)
［8］	Zhang Shaozhong, Feng Yingjun. Robustness of SPDS algorithm for BP network ［J］. Journal of Harbin Institute of
	Table 2Training results of SPDS and refrence［9］ about geological data analysis表2地质数据分析的SPDS算法训练结果与文献［9］的结果对比编号SiO2CaOMgOAl2O3K2ONa2OMoSPDS
	训练结果线性回归
结果［7］	173.740.330.8210.755.200.580.0770.0690.101272.730.581.1411.975.480.890.0700.0780.131372.441.621.4510.884.800.140.2270.2270.186477.420.520.548.634.360.400.0430.0470.065580.420.220.717.864.130.220.0450.0560.070682.790.200.316.633.400.160.0410.0250.016776.230.520.569.725.380.360.0620.0780.107877.421.430.708.224.500.080.0730.0760.123976.630.770.829.264.430.530.1070.0960.0911072.931.100.9310.595.380.300.2290.2210.1501172.020.711.0111.855.381.250.1570.1530.1081271.281.371.0511.504.830.360.2220.2230.1401368.101.660.9014.664.104.010.0110.005-0.0231470.861.300.1413.615.202.320.0070.0100.0171568.511.500.8914.294.263.160.0050.0050.0071666.442.081.1213.764.042.050.0170.0210.0691768.961.280.8014.114.192.860.0190.0050.0021867.121.000.7914.614.132.710.003-0.002-0.0081971.081.020.5113.295.351.570.006-0.0010.0672068.461.980.9314.133.943.740.0050.006-0.0072167.382.060.8813.433.754.010.0000.006-0.0272266.542.201.3515.193.483.860.0220.0050.0052365.631.481.7615.223.634.010.0000.0090.0112464.892.541.3515.503.194.040.0000.005-0.0052569.201.950.5613.763.502.540.0000.011-0.0102661.983.191.9515.163.083.720.0010.0080.063Table 3Training results of neural network (weights and thresholds)表3网络训练结果（阈值和权值）阈值输入层
	单元一权值输入层
	单元二权值输入层
	单元三权值输入层
	单元四权值输入层
	单元五权值输入层
	单元六权值隐层单元一-2.330e+009.602e-03 -3.613e+005.826e+00-3.320e-011.063e-01-1.613e+00隐层单元二-2.811e+00-2.360e-011.668e-011.216e-01-2.988e-023.588e-01-1.614e+00隐层单元三-2.865e+001.090e-01 -5.638e+004.740e+00-4.330e-01 -4.319e-01-7.465e+00隐层单元四-5.081e-012.699e-011.369e+001.256e+000.000e+000.000e+000.000e+00隐层单元五0.000e+002.908e-01 6.403e-020.000e+001.202e-010.000e+000.000e+00隐层单元六8.735e-01-1.633e-01-6.285e-011.239e+006.409e-015.523e-01-1.882e+00阈值隐层
	单元一权值隐层
	单元二权值隐层
	单元三权值隐层
	单元四权值隐层
	单元五权值隐层
	单元六权值输出层单元-3.156e-022.252e-011.168e+05-1.954e-01-2.716e-021.355e-03-3.092e+00Technology, 2000, 32(2):6061. (in Chinese)
［9］	Xu Z B,Lou Y R,Jin C H.All regression analysis and optimization method and the application in geology［M］. Beijing:Geology Press, 1990.（in Chinese）
［10］	Hornik K. Some new results on neural network approximation ［J］. Neural Networks, 1993, 4(6):10691072.
	附中文参考文献:
［1］	张铃，张钹. 神经网络中BP算法的分析［J］. 模式识别与人工智能，1994，7 (3):191195.
［2］	王雪峰，冯英浚. 多层神经网络的一种新的学习算法［J］. 哈尔滨工业大学学报，1997, 29 (2):2325.
［5］	冯英浚, 张少仲, 王雪峰, 等. BP网络瘫痪的原因分析［J］. 哈尔滨工业大学学报, 2000, 32 (5):8183.
［6］	张少仲, 冯英浚, 王雪峰, 等. 基于单参数坐标轮换法的多层前馈型神经网络的一种新型算法［J］. 电子学报, 1999, 27 (2):115116.
［7］	张少仲, 冯英浚. 单参数动态搜索算法的泛化与容错能力［J］. 哈尔滨工业大学学报, 2000, 32 (6):14.
［8］	张少仲, 冯英浚. BP网络的SPDS训练算法的鲁棒性［J］. 哈尔滨工业大学学报, 2000, 32 (2):6061.
［9］	徐振邦，娄元仁，金淳浩.全部回归分析与优选方法及其在地质学中的应用［M］. 北京：地质出版社，1990.

[1]	张晓双1，夏群峰2，刘渊1，徐雁飞1. 社交网络用户影响力分析ABP算法研究与应用[J]. 计算机工程与科学, 2017, 39(03): 477-484.
[2]	张志文，乔悦，罗隆福，杨双. 信息融合技术在变压器油气识别故障诊断中的研究[J]. J4, 2012, 34(3): 132-136.
[3]	杨灿群，杨学军，易会战. 扩展双精度浮点并行计算：MPI方法[J]. J4, 2010, 32(12): 98-101.
[4]	秦焱朱宏李旭伟. 基于改进型粒子群优化算法的BP网络在股票预测中的应用[J]. J4, 2008, 30(4): 66-68.
[5]	刘兰兰朱双东. 两种智能交通标志分类器的比较研究[J]. J4, 2007, 29(2): 62-65.
[6]	罗莉[1] 罗强[2] 等. 前馈多层神经网络BP算法与可靠性增长模型[J]. J4, 2001, 23(3): 55-58.
[7]	张铃[1] 张钹[2]. 前向神经网络设计问题的回顾与探索[J]. J4, 1998, 20(4): 1-10.
[8]	柳文[1] 冯建农[2]. 一种改进型快速BP训练算法[J]. J4, 1998, 20(4): 17-20.