基于神经网络方法在致密砂岩可钻性中的建模
|
摘要
随着石油工业的发展和勘探开发的进行，致密砂岩气藏勘探开发工作越来越受到人们的关注，中国致密气资源极为丰富，第三次油气资源评价显示，致密气资源量约12×10~(12)m~3，占天然气总资源量40％左右，这些致密气主要赋存于四川、陕甘宁、松辽、塔里木、准噶尔和柴达木盆地。川中地区上三叠统香溪群埋深2100～～3600m，厚度450～～1000m，为一套自生自储巨厚砂体，其中主力产气层香四、香二段岩性主要为灰色、灰白色中粒长石石英砂岩、中粒岩屑长石石英沙岩。储层孔隙度分布范围在1.39％～～12.22％，平均孔隙度5.71％，渗透率分布在(0.001～～0.953)×10~(-3)um~2，平均渗透率为0.064×10~(-3)um~3。压汞资料分析表明，最大喉道半径为0.77～～2.42um，饱和中值喉道半径为0.06～～0.387um，其中50％的喉道半径在0.25 um以下。该储层为典型的致密砂岩气藏。
本文以川西须家河组致密砂岩为研究对象。选取该地区的川孝新560井作为标准井，结合该井的录井资料建立起该地区的致密砂岩可钻性径向基神经网络模型。并对该地区的最新855井致密砂岩层进行了可钻性值预测。本文在以下几个方面做了较深入的探讨：
1)综合国内外岩石可钻性评价的现状，分析了目前现阶段的主要成果以及不足，得到了目前传统的可钻性评价方法还不太适合于致密砂岩可钻性评价，而且各方技术人员都在积极探索评价致密性岩储层物性的有效方法，准确评价致密性岩储层物性已成为目前天然气勘探亟待解决的技术难题。
2)考虑引进非线性多元回归解决致密性岩可钻性的研究，系统阐述了非线性回归的理论，方法以及一些常见的非线性回归模型，最后通过分析提出了用非线性回归解决致密性岩可钻性研究的难题及解决方法。
3)用神经网络方法评价致密砂岩可钻性：运用RBF神经网络结合录井资料，通过对致密砂岩可钻性相关因素的分析，采用RBF网络模型进行拟合和预测。依据川西须家河组的钻井数据进行网络学习训练和预测，得到还较好的结果。
With development of oil industry and the exploration and development being in progress, the compact sandstone gas Tibet exploration and development job catches people's attention, compact Chinese gas resource enriches extremely, third time oil and gas resources appraises display, compact gas resource amounts roughly account for about 40% of general gas resource amounts, these Compact gas main gas tax exists in Sichuan, Shanxi, Gansu and Ningxia, loose Liaoning, Talimu, Sungar and the Qaidam Basin. Triple-lap commands the incense streamlet in central Sichuan on area burying 2100～～3600 deep ms, thickness group 450～～1000 ms, the tier is fragrant for a set of been born in by self and stored the huge favor the grit body, among them main force up by self produces gas four, incense two section of rock sex is that the gray, the off-white are hit by granule feldspar quartz sandstone, are hit by granule rock fragments feldspar quartz sandrock mainly. The reservoir small opening distributes range degree 12.22%, shares a small opening spending 5.71%, penetrance distribution is 0.064×10~(-3)um~3 in average penetrance in 1.39%. Data analysis indicates the pressure mercury, the maximal throat says that the radius is 0.77～～2.42um, the value throat says that the radius is 0.06～～0.387um in saturation, 50%'s throat says that the radius is in 0.25um following among them. Be a reservoir's turn to conceal self for representative compact sandstone gas.
With west Sichuan beard home, the river forms the main body of a book for compact sandstone studying a marriage partner. Choose owe the area river filial piety 560 new wells action the standard well, owing compact area sandstone but drill nature radial direction base nerve network model combining with being the well record well data building-up's tum to get up. And, the value tier, having been in progress but studying nature intensively to owing 855 up-to-date area wells compact sandstone forecasts. Under the main body of a book being in, several aspect has made more thorough investigation and discussion:
1) The rock may drill the current situation that nature appraises synthetically at home and abroad, analytical main the present stage achievement has been at present as well as insufficient, the tradition may studies nature intensively at present appraising method fairly not very suitable than compact sandstone may drill nature valuation, gas prospects the and respectively square technician at present all in trying to explore new ways estimating that effective compactness rock reservoir thing nature method, accurate valuation compactness rock reservoir thing nature already have become waiting for the technology difficult problem solving urgently.
2) Research thinking that the rock resolving compactness may study nature intensively introducing multivariant return of nonlinearity, system has set forth nonlinearity return theory, method has analysed the problem and solution having brought forward nonlinearity return as well as a few common nonlinearity return model, passes finally.
3) With neural networks, method appraises compact sandstone but studies nature intensively: The neural networks wielding RBF being recorded the well data combining with, analysed by may study the nature relevance factor's intensively to compact sandstone, adopt the RBF network model to carry out a fitting and forecast. Learn training and forecasting according to the borehole of west Sichuan beard family river group data carries out a network, get fairly good result.
本文以川西须家河组致密砂岩为研究对象。选取该地区的川孝新560井作为标准井，结合该井的录井资料建立起该地区的致密砂岩可钻性径向基神经网络模型。并对该地区的最新855井致密砂岩层进行了可钻性值预测。本文在以下几个方面做了较深入的探讨：
1)综合国内外岩石可钻性评价的现状，分析了目前现阶段的主要成果以及不足，得到了目前传统的可钻性评价方法还不太适合于致密砂岩可钻性评价，而且各方技术人员都在积极探索评价致密性岩储层物性的有效方法，准确评价致密性岩储层物性已成为目前天然气勘探亟待解决的技术难题。
2)考虑引进非线性多元回归解决致密性岩可钻性的研究，系统阐述了非线性回归的理论，方法以及一些常见的非线性回归模型，最后通过分析提出了用非线性回归解决致密性岩可钻性研究的难题及解决方法。
3)用神经网络方法评价致密砂岩可钻性：运用RBF神经网络结合录井资料，通过对致密砂岩可钻性相关因素的分析，采用RBF网络模型进行拟合和预测。依据川西须家河组的钻井数据进行网络学习训练和预测，得到还较好的结果。
With development of oil industry and the exploration and development being in progress, the compact sandstone gas Tibet exploration and development job catches people's attention, compact Chinese gas resource enriches extremely, third time oil and gas resources appraises display, compact gas resource amounts roughly account for about 40% of general gas resource amounts, these Compact gas main gas tax exists in Sichuan, Shanxi, Gansu and Ningxia, loose Liaoning, Talimu, Sungar and the Qaidam Basin. Triple-lap commands the incense streamlet in central Sichuan on area burying 2100～～3600 deep ms, thickness group 450～～1000 ms, the tier is fragrant for a set of been born in by self and stored the huge favor the grit body, among them main force up by self produces gas four, incense two section of rock sex is that the gray, the off-white are hit by granule feldspar quartz sandstone, are hit by granule rock fragments feldspar quartz sandrock mainly. The reservoir small opening distributes range degree 12.22%, shares a small opening spending 5.71%, penetrance distribution is 0.064×10~(-3)um~3 in average penetrance in 1.39%. Data analysis indicates the pressure mercury, the maximal throat says that the radius is 0.77～～2.42um, the value throat says that the radius is 0.06～～0.387um in saturation, 50%'s throat says that the radius is in 0.25um following among them. Be a reservoir's turn to conceal self for representative compact sandstone gas.
With west Sichuan beard home, the river forms the main body of a book for compact sandstone studying a marriage partner. Choose owe the area river filial piety 560 new wells action the standard well, owing compact area sandstone but drill nature radial direction base nerve network model combining with being the well record well data building-up's tum to get up. And, the value tier, having been in progress but studying nature intensively to owing 855 up-to-date area wells compact sandstone forecasts. Under the main body of a book being in, several aspect has made more thorough investigation and discussion:
1) The rock may drill the current situation that nature appraises synthetically at home and abroad, analytical main the present stage achievement has been at present as well as insufficient, the tradition may studies nature intensively at present appraising method fairly not very suitable than compact sandstone may drill nature valuation, gas prospects the and respectively square technician at present all in trying to explore new ways estimating that effective compactness rock reservoir thing nature method, accurate valuation compactness rock reservoir thing nature already have become waiting for the technology difficult problem solving urgently.
2) Research thinking that the rock resolving compactness may study nature intensively introducing multivariant return of nonlinearity, system has set forth nonlinearity return theory, method has analysed the problem and solution having brought forward nonlinearity return as well as a few common nonlinearity return model, passes finally.
3) With neural networks, method appraises compact sandstone but studies nature intensively: The neural networks wielding RBF being recorded the well data combining with, analysed by may study the nature relevance factor's intensively to compact sandstone, adopt the RBF network model to carry out a fitting and forecast. Learn training and forecasting according to the borehole of west Sichuan beard family river group data carries out a network, get fairly good result.
引文
[1] 刘向君，孟英峰．岩石可钻性和钻速预测．天然气工业，1999；19(5)：61～63．
[2] 刘希圣等编．钻井工艺原理．北京：石油工业出版社，1981．
[3] 何龙，朱澄清．川西地层可钻性级值研究．钻采工艺，2006，29(3)：98—99．
[4] 卢春阳，等．盐城地区可钻性分级及钻头选型[J] ．钻采工艺，2001，24(6)：10—12．
[5] HOERL A E, KENNARD R W. Ridge Regession Biased Estrmation for Non-orthogonal Problem[J] . Techncmetrics, 1970, 12: 55-88.
[6] C A. Micchelli, "Interpolation of scattered data: distance matrices and conditionally positive definite functions", Contructive Approximation, Vol. 2, pp. 1 1-22, 1986.
[7] M. I. D. Powell, "Radial basis functions for multivariable interpolation: a review", IMA Conf. on Algorithms for the Approximation of Functions and Data, RMCS Shrivenham, 1985.
[8] M. I. D. Powell, "Radial basis function approximations to polynomials", 12th Biennial Numerical Analysis Con#, Dundee, pp. 232-241, 1987.
[9] D. S. Broomhead and D. Lowe, "Multivariable functional interpolation and adaptive networks", Complex Systems, Vol. 2, pp. 321-355, 1988.
[10] S. Chen, S. A. Bilings, C. F. N. Cowan, and P. M. Grant, "Non-linear systems identifi -cation using radial basis functions", International Journal of Systems Science, Vol. 2. 1, pp. 2513-2539, 1990.
[11] S. Chen, S. A. Bilings, C. F. N. Cowan, and P. M. Grant, "Practical identification of NARMAR models using radial basis functions", International Journal of Control, Vol. 52, pp. 1327—1350, 1990
[12] 阎平凡、张长水编，人工神经网络与模拟进化计算，北京：清华大学出版，2000．
[13] Simon Haykin, NEURAL NETWORKS-A Comprehensive Foundation, Prentice Hall, Second Edition.
[14] J. Barry Gomma and Ding Li Tu, "Selecting Radial Basis Function Centers with Recursive Orthogonal Least Squares Training, " IEEE Tras. on NN. , vol. 11, 2000.
[15] Zekeriya U and C. G, M. Ertugrul. C and t-Ieikki N. Koivo, "Analysis of Input-Output Clustering for Determining Centers of RBFN, " IEEE Tras, on NN, vol. 4(5), pp. 851-858, 2000.
[16] W. Kaminski and P. Strumillo, "Kernel Orthonormalization in Radial Basis Function Neural Networks, "IEEE Tras, on NN, vol. 4(5), pp. 1177-1183, 1997.
[17] Xu Letal, "Rival Penalized Competitive Learning for Clustering Analysis RBF Net and Curve Detection, " IEEE Tras, on NN, vol. 4(4), pp. 636-649, 1993.
[18] Nikhil R. pal, "Generalized Clustering Networks and Kohonen's Self-Organizing Scheme, " IEEE Tras, on NN, vol. 4(4), pp. 549-556, 1993.
[19] 王永骥，涂健．神经儿网络控制[M] ．北京：机械土业出版社，1998．
[20] 阎平凡，张长水．人土神经网络与模拟进化计算[M] ．北京：清华大学出版社，2000．
[21] 周纪芗，《回归分析》，华东师范大学出版社，1993．
[22] 吴承广，马良河，《幕指函数回归模型的几个应用》，《数理统计与管理》，1996年第5期。
[23] Harrison, P. J, and Stevens, C. F. , A Bayesian approach to short-term Forecasting oper. Res. Qurat 22(1971), 341-362.
[24] Warwick university Hayesian Forecasting Group Training Programme. FAB.
[25] Harrison, P. J, and West, M. , Bayesian Forecasting and Dynamic Models, Springer-VerIag, New York (1989).
[26] 洪再吉、韦博成等译，非线性回归模型，南京大学出版社，1986．
[27] 王宪举、金茎，一类非线性模型的稳健回归，数学的实践与认识，1(1987)，31-35．
[28] Stephen. A. Holditch. Factors Affecting Water Blocking and Gas Flow From Hydraulically Fractured Gas Wells. JPT, 1979: 1515-1524.
[29] Moody. J. E, Darken. C. J. Fast Learning in Networks of Locally-turned Processing Units[J] . Neural Computation. 1989, 1: 281-294.
[30] Chen S. Billing S A. Grant P M. Recursive Hybrid Algorithm for Identification Using Radial Basis Function Networks. International Journal of Control, 1995, 55(5): 1051-1070.
[31] Chen. T, Chen. H, Liu RW. Approximation Capacity in C(Rn) by Multilayer Feedforward Networks and Related Problems. IEEE Transon Neural Networks, 1995, 6(1).
[32] Moshe Leshno, etal. Multilayer Feedforward Networks with Nonpolynomial Activation Function Can Approximate Any Function Neural Networks, 1993, 6: 861-867.
[33] Honik K. Some New Results on Neural Network Approximation. Neural Networks 1993, 6: 1069-1072.
[34] J Park, Sandberg I W. Universal Approximation Using Radial Basis Function Networks. Neural Computation, 1991, 3: 246-257.
[35] Y-h. Luo and S-y Shen, L~p Approximation of Sigma-Pi Neural Networks, SPIE, 2825 (1996) 152-162.
[36] 吴微，神经网络计算，高等教育出版社，2003
[37] 刘志远，吕剑虹，陈来九．基于RBF神经网络的单儿机组负荷系统建模研究．控制与决策．2003，185)：637-640．
[38] 刘兴华，刘宪英，胡泽．基于RBF神经网络系统辨识研究．现代电子技术．2004(24)：57—59．
[39] 宋宜斌，王培进．基于径向基函数神经网络的非线性模型辨识．计算机工程．2004，30(5)：142—143．
[40] M．S 博格著，余庆余译，非线性与泛函分析，科学出版社．1989．
[41] 邓星亮，吴清．LLE算法及其应用[J] ．测控技术，2005，24(3)：65-68．
[42] 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．
[43] 杨建刚．人工神经网络实用教程[M] ．杭州：浙江大学出版社，2002．
[44] 赵永刚．人工神经网络中BP算法及其改进的思考[J] ．广东财经职业学院学报，2003，2(4)：79-82．
[45] 徐海银等．容错神经网络及其容错BP算法研究[J] ．华中理工大学学报，1995，23(2)：345-347．
[46] 宋宜斌，王培进．多层前馈神经网络改进算法及其应用[J] ．计算机工程，2003，29(14)：109-112．
[47] 刘洪林，胡建．GA+BP神经网络在薄互层储层预测中的应用[J] ．物探与化探，2004，28(5)：460-463．
[48] 刘洪波，王秀坤等．神经网络基于粒子群优化的学习算法研究[J] ．小型微型计算机系统，2005，26(4)：638-640
[49] Rosipal. R, Koska. M and Farkas. I, "Chaotic time-series prediction using resource-allocating RBF networks, "Interna ti oval conference aleasurement'97, Smolenice, 1997.
[50] Rosipal. R, Koska. M and Farkas. I, "Prediction of Chaotic TimeSeries with a Resource-Allocating RBF Network, " Neural Processing Letters 7, pp. 185-197, 1998.
[51] Wang. Y, Guang-Bin Huang, "Self-adjustment of neuron impact widthin growing and pruning RBF (GAP-RBF) neuron networks, " Information, Communications and Signal Processing, and the Fourth Pacific Rim Conference on Multimedia. Proceedings Of the 2003 Joint Conference of the Fourth International Conference on, vol. 2, pp14-1017, 2003.
[52] 李运华等，“神经网络离散变结构控制的算法及其应用”，控制与决策．Vol．9，No．4，pp，301-305，1994．
[53] Martin T Hagan，Howard B Demuth，Mark H Beale．戴葵等译．神经网络设计[M] ．北京：机械工业出版社，2002．9．
[54] 飞思科技产品研发中心．MATLAB6. 5辅助神经网络分析与设计[M] ．北京：电子工业出版社，2003．1．
[55] 郭科，龚灏．多元统计方法及其应用[M] ．成都：电子科技大学出版社，2003．
[2] 刘希圣等编．钻井工艺原理．北京：石油工业出版社，1981．
[3] 何龙，朱澄清．川西地层可钻性级值研究．钻采工艺，2006，29(3)：98—99．
[4] 卢春阳，等．盐城地区可钻性分级及钻头选型[J] ．钻采工艺，2001，24(6)：10—12．
[5] HOERL A E, KENNARD R W. Ridge Regession Biased Estrmation for Non-orthogonal Problem[J] . Techncmetrics, 1970, 12: 55-88.
[6] C A. Micchelli, "Interpolation of scattered data: distance matrices and conditionally positive definite functions", Contructive Approximation, Vol. 2, pp. 1 1-22, 1986.
[7] M. I. D. Powell, "Radial basis functions for multivariable interpolation: a review", IMA Conf. on Algorithms for the Approximation of Functions and Data, RMCS Shrivenham, 1985.
[8] M. I. D. Powell, "Radial basis function approximations to polynomials", 12th Biennial Numerical Analysis Con#, Dundee, pp. 232-241, 1987.
[9] D. S. Broomhead and D. Lowe, "Multivariable functional interpolation and adaptive networks", Complex Systems, Vol. 2, pp. 321-355, 1988.
[10] S. Chen, S. A. Bilings, C. F. N. Cowan, and P. M. Grant, "Non-linear systems identifi -cation using radial basis functions", International Journal of Systems Science, Vol. 2. 1, pp. 2513-2539, 1990.
[11] S. Chen, S. A. Bilings, C. F. N. Cowan, and P. M. Grant, "Practical identification of NARMAR models using radial basis functions", International Journal of Control, Vol. 52, pp. 1327—1350, 1990
[12] 阎平凡、张长水编，人工神经网络与模拟进化计算，北京：清华大学出版，2000．
[13] Simon Haykin, NEURAL NETWORKS-A Comprehensive Foundation, Prentice Hall, Second Edition.
[14] J. Barry Gomma and Ding Li Tu, "Selecting Radial Basis Function Centers with Recursive Orthogonal Least Squares Training, " IEEE Tras. on NN. , vol. 11, 2000.
[15] Zekeriya U and C. G, M. Ertugrul. C and t-Ieikki N. Koivo, "Analysis of Input-Output Clustering for Determining Centers of RBFN, " IEEE Tras, on NN, vol. 4(5), pp. 851-858, 2000.
[16] W. Kaminski and P. Strumillo, "Kernel Orthonormalization in Radial Basis Function Neural Networks, "IEEE Tras, on NN, vol. 4(5), pp. 1177-1183, 1997.
[17] Xu Letal, "Rival Penalized Competitive Learning for Clustering Analysis RBF Net and Curve Detection, " IEEE Tras, on NN, vol. 4(4), pp. 636-649, 1993.
[18] Nikhil R. pal, "Generalized Clustering Networks and Kohonen's Self-Organizing Scheme, " IEEE Tras, on NN, vol. 4(4), pp. 549-556, 1993.
[19] 王永骥，涂健．神经儿网络控制[M] ．北京：机械土业出版社，1998．
[20] 阎平凡，张长水．人土神经网络与模拟进化计算[M] ．北京：清华大学出版社，2000．
[21] 周纪芗，《回归分析》，华东师范大学出版社，1993．
[22] 吴承广，马良河，《幕指函数回归模型的几个应用》，《数理统计与管理》，1996年第5期。
[23] Harrison, P. J, and Stevens, C. F. , A Bayesian approach to short-term Forecasting oper. Res. Qurat 22(1971), 341-362.
[24] Warwick university Hayesian Forecasting Group Training Programme. FAB.
[25] Harrison, P. J, and West, M. , Bayesian Forecasting and Dynamic Models, Springer-VerIag, New York (1989).
[26] 洪再吉、韦博成等译，非线性回归模型，南京大学出版社，1986．
[27] 王宪举、金茎，一类非线性模型的稳健回归，数学的实践与认识，1(1987)，31-35．
[28] Stephen. A. Holditch. Factors Affecting Water Blocking and Gas Flow From Hydraulically Fractured Gas Wells. JPT, 1979: 1515-1524.
[29] Moody. J. E, Darken. C. J. Fast Learning in Networks of Locally-turned Processing Units[J] . Neural Computation. 1989, 1: 281-294.
[30] Chen S. Billing S A. Grant P M. Recursive Hybrid Algorithm for Identification Using Radial Basis Function Networks. International Journal of Control, 1995, 55(5): 1051-1070.
[31] Chen. T, Chen. H, Liu RW. Approximation Capacity in C(Rn) by Multilayer Feedforward Networks and Related Problems. IEEE Transon Neural Networks, 1995, 6(1).
[32] Moshe Leshno, etal. Multilayer Feedforward Networks with Nonpolynomial Activation Function Can Approximate Any Function Neural Networks, 1993, 6: 861-867.
[33] Honik K. Some New Results on Neural Network Approximation. Neural Networks 1993, 6: 1069-1072.
[34] J Park, Sandberg I W. Universal Approximation Using Radial Basis Function Networks. Neural Computation, 1991, 3: 246-257.
[35] Y-h. Luo and S-y Shen, L~p Approximation of Sigma-Pi Neural Networks, SPIE, 2825 (1996) 152-162.
[36] 吴微，神经网络计算，高等教育出版社，2003
[37] 刘志远，吕剑虹，陈来九．基于RBF神经网络的单儿机组负荷系统建模研究．控制与决策．2003，185)：637-640．
[38] 刘兴华，刘宪英，胡泽．基于RBF神经网络系统辨识研究．现代电子技术．2004(24)：57—59．
[39] 宋宜斌，王培进．基于径向基函数神经网络的非线性模型辨识．计算机工程．2004，30(5)：142—143．
[40] M．S 博格著，余庆余译，非线性与泛函分析，科学出版社．1989．
[41] 邓星亮，吴清．LLE算法及其应用[J] ．测控技术，2005，24(3)：65-68．
[42] 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．
[43] 杨建刚．人工神经网络实用教程[M] ．杭州：浙江大学出版社，2002．
[44] 赵永刚．人工神经网络中BP算法及其改进的思考[J] ．广东财经职业学院学报，2003，2(4)：79-82．
[45] 徐海银等．容错神经网络及其容错BP算法研究[J] ．华中理工大学学报，1995，23(2)：345-347．
[46] 宋宜斌，王培进．多层前馈神经网络改进算法及其应用[J] ．计算机工程，2003，29(14)：109-112．
[47] 刘洪林，胡建．GA+BP神经网络在薄互层储层预测中的应用[J] ．物探与化探，2004，28(5)：460-463．
[48] 刘洪波，王秀坤等．神经网络基于粒子群优化的学习算法研究[J] ．小型微型计算机系统，2005，26(4)：638-640
[49] Rosipal. R, Koska. M and Farkas. I, "Chaotic time-series prediction using resource-allocating RBF networks, "Interna ti oval conference aleasurement'97, Smolenice, 1997.
[50] Rosipal. R, Koska. M and Farkas. I, "Prediction of Chaotic TimeSeries with a Resource-Allocating RBF Network, " Neural Processing Letters 7, pp. 185-197, 1998.
[51] Wang. Y, Guang-Bin Huang, "Self-adjustment of neuron impact widthin growing and pruning RBF (GAP-RBF) neuron networks, " Information, Communications and Signal Processing, and the Fourth Pacific Rim Conference on Multimedia. Proceedings Of the 2003 Joint Conference of the Fourth International Conference on, vol. 2, pp14-1017, 2003.
[52] 李运华等，“神经网络离散变结构控制的算法及其应用”，控制与决策．Vol．9，No．4，pp，301-305，1994．
[53] Martin T Hagan，Howard B Demuth，Mark H Beale．戴葵等译．神经网络设计[M] ．北京：机械工业出版社，2002．9．
[54] 飞思科技产品研发中心．MATLAB6. 5辅助神经网络分析与设计[M] ．北京：电子工业出版社，2003．1．
[55] 郭科，龚灏．多元统计方法及其应用[M] ．成都：电子科技大学出版社，2003．