(3.238.235.155) 您好!臺灣時間:2021/05/16 07:31
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果

詳目顯示:::

: 
twitterline
研究生:陳悅明
研究生(外文):Yue-Ming Chen
論文名稱:使用SupportVectorRegression建構複雜網路系統之反應曲面模型
論文名稱(外文):Modeling the Response Surface for Complex Queueing Networks using Support Vector Regression
指導教授:洪英超
指導教授(外文):Ying-Chao Hung
學位類別:碩士
校院名稱:國立中央大學
系所名稱:統計研究所
學門:數學及統計學門
學類:統計學類
論文種類:學術論文
論文出版年:2005
畢業學年度:93
語文別:英文
論文頁數:31
外文關鍵詞:Generalized Switch Modeltreed modelMaximum Weighted Queue Length policyMaximum Service Rate policyMultivariate Adaptive Regression SplinesCubic Smoothing SplinesSupport Vector Regression
相關次數:
  • 被引用被引用:0
  • 點閱點閱:278
  • 評分評分:
  • 下載下載:44
  • 收藏至我的研究室書目清單書目收藏:2
  目前網路分析的重點在透過系統反應的表現值來研究系統的動態關係。一些重要的系統表現值包括:吞吐量(Throughput)、延遲時間(Delay)和存貨(Backlog)等等。由於現代網路系統的複雜度與日俱增,系統的表現值通常受到大量輸入參數的影響,所以理論分析的結果經常是不易取得的。在這種情況下,電腦模擬(Computer Simulation) 便成為分析複雜網路系統的重要工具。為了了解輸入參數和系統反應值之間的關係,文獻上傳統的作法是利用電腦模擬在整個輸入參數空間裡建構一個有母數模型 (Parametric Model)。這種作法的好處是我們可以清楚地知道輸入參數如何影響系統反應值的大小,但是這類作法在高維度的輸入空間也同時遭遇到模型選擇的問題。因此,當我們對複雜系統的反應所知不多時,使用無母數 (Nonparametric) 的方法似乎較為適當。

  此論文介紹一無母數方法來建構複雜網路系統的反應曲面模型。其目的是(i)希望用比較少的模擬次數得到不錯的反應曲面模型;(ii)希望所提出的方法可以較容易處理高維度參數空間的問題。本文也介紹一個被稱為廣泛交換系統(Generalized Switch Model)的網路模型,並以此模型示範我們所建議的方法。
The goal of network analysis has been focused on studying the dynamics of a system through
important performance measures such as throughput, delay, backlog and so on. Due to the
significant increase on the complexity of modern networks, the performance measures are
usually affected by a lot of input parameters, thus analytical solutions are often invalid.
Therefore, one often relies on simulation when analyzing complex network systems. Typically,
a parametric model is built over the entire input space so that the relationship between the
response measures of interest and the input parameters can be well described. However,
parametric methods suffer from the issues like model selection, computational validity, etc.
Therefore, non-parametric methods seem to be more plausible in analyzing complex network
systems when prior information is not valid. The goal of this study is to find an adequate
non-parametric method so that a good model for the response surface can be built using
a possibly smaller number of simulation runs and the model can also perform well in high-dimensional
input space. Among all non-parametric methods, support vector regression (SVR)
is considered in this study. This is mainly due to the following two reasons. First, it might
request fewer simulation runs than other approaches. Second, it can easily deal with high-dimensional
input spaces. A particular queueing system called the generalized switch model
is introduced and used to demonstrate the proposed approach.
1 Introduction 1
2 A Generalized Switch Model 4
3 Support Vector Regression 7
3.1 Risk Function 7
3.2 Example 8
3.3 Kernels 10
4 Apply SVR for Generalized Switch Models 12
4.1 Model Construction 12
4.1.1 The Average Sojourn Time Surfaces 14
4.2 Comparison with Other Approaches via Predictions 17
4.3 Ad Hoc Applications 19
4.3.1 Compare the Average-Sojourn-Time Surfaces for Two Di erent Sets of
Servers 19
4.3.2 Compare the Average-Sojourn-Time Surfaces for Two Di erent Control
Policies 21
5 Conclusion and Future Work 23
Bibliography 24
[1] Alexander, W. P. and S. D. Grimshaw. Treed regression. Journal of Computational and
Graphical Statistics 5 (1996), pages 156-175.
[2] Hung, Y. C. Modeling and analysis of stochastic networks with shared resources. Ph.D.
thesis, Department of Statistics, The University of Michiganm, Ann Arbor, Michigan.
2002.
[3] Hung, Y. C., Michailidis, G. and Bingham, D. R.. Developing E cient Simulation
Methodology for Complex Queueing Networks. Proceedings of the Winter Simulation Conference,
New Orlean. pages 152-159, 2003.
[4] Alex J. Smola and Bernhard Scholkopf. A Tutorial on Support Vector Regression. September
30, 2003.
[5] Nello Cristianini and John Shawe-Taylor. An Introduction to Support Vector Machines
and ither kernel-based learning methods. Cambridge University Press, 2000.
[6] Vladimir N. Vapnik. The nature of Statistical Learning Theory. New York: Springer,
1995.
[7] P. J. Green and B. W. Silverman. Nonparametric Regression and Generalized Linear
Models: A roughness penalty approach. Chapman & Hall. 1994.
[8] C. J. Stone, M. Hansen, C. Kooperberg and Y. K. Truong. Polynomial Splines and their
tensor products in extended linear modeling. Annals of Statistics, 25 (1997), pages 1371-
1470.
[9] Jerome H. Friedman. Multivariate Adaptive Regression Splines. Annals of Statistics, 19
(1991), pages 1-67.
[10] W. N. Venables and B. D. Ripley. Modern Applied Statistics with S, 4th Edition. New
York:Springer, 2002.
[11] Kai-Tai Fang and Dennis K. J. Lin. Uniform Experimental Designs and their Applications
in Industry. Handbook in Statistics: Statistics in Industry, 2003.
[12] 張惠敏。設計複雜網路系統之高效率模擬方法。碩士論文,統計研究所,國立中央大學,中壢,台灣。2004。
[13] 邱啟宗。可資源共享之平行分散系統的最大吞吐量控制策略。碩士論文,統計研究所,國立中央大學,中壢,台灣。2004。
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top