臺灣博碩士論文加值系統

於台灣位處於熱帶與亞熱帶地區，特殊的地理位置使得遭受熱帶氣旋(Tropical Cyclone)也就是颱風侵襲的頻率相當高。而目前中央氣象局防颱應變中心將過去有發佈警報之颱風侵台路徑區分為九群，本研究提出的一個藉由資料探勘中的分群演算法來做即時的颱風路徑預測。
本研究目的在於藉由歷史的侵台颱風資料，建立一分群預測模型：CTPA(Clustering-based Trajectory Prediction Algorithm)分群路徑向量演算法架構
，模型採用1996年至2008年中央氣象局發佈警報之颱風路徑其經、緯度為訓練資料集，利用最大期望值演算法(EM algorithm)及適應式模糊網路模型(ANFIS)建立預測模型，並以2009及2010年發佈警告之颱風做驗證，實驗結果顯示：雖使用少數屬性，使用CTPA預測颱風路徑演算法可趨近颱風的實際路徑。
本研究僅建立一種颱風預測之雛型，藉由所獲颱風資料屬性之增加，可增加颱風預測之準確率。希望能透過此預測模式能提供防颱中心與人民防災之參考，並能即時預測侵台颱風路徑，提早完成防災措施，減低天然災害所造成的生命財產損失。

Since Taiwan is located in the tropical and subtropical zones, several typhoons could be generated in every single year. In this special geographical location, it is with high probability that these typhoons attack Taiwan. The TCWB (Taiwan Central Weather Bureau) classified the history typhoons that attacked Taiwan into nine classes. In this study, we take advantage of various data mining techniques to create a real-time typhoon trajectory forecasting model, Clustering-based Trajectory Prediction Algorithm (abbreviated to CTPA).
CTPA uses Expectation-Maximum algorithm for building a clustering model to obtain a preliminary typhoon trajectory, and then an adaptive-network-based fuzzy inference system, ANFIS, is used to refine the trajectory.
In this study, typhoons from 1996 to 2008 released by TCWB are used as training data while those from 2009 to 2010 are used as test data. Experimental results show that trajectories predicted by CTPA achieving high accuracy compared to the respective real typhoon trajectories.
CTPA is a prototype model for typhoon trajectory prediction, which just uses a few attributes and is able to achieve some accuracy in real time. Higher accuracy and more approximate trajectory forecasting can be obtained, if more attributes could be collected and used to train the CTPA model. We hope that the model and the results developed in this study could be a useful reference for TCWB and other meteorological personnel.

誌謝 i
摘要 ii
ABSTRACT iii
目錄 iv
表目錄 vi
圖目錄 viii
1. 緒論 1
1.1研究動機 1
1.2研究目的 1
1.3研究方法 1
1.4論文架構 1
2. 文獻探討 3
2.1馬克夫鏈模擬法 4
2.2支援向量機 4
2.2.1線性支援向量機 4
2.2.2非線性支援向量機 5
2.3最大期望值分群法 6
2.3.1高斯混合模型 8
2.4線性迴歸 10
2.4.1 多項式迴歸 10
2.4.2最小平方法 11
2.5調適性網路模糊推論系統 12
2.5.1類神經網路 14
2.5.2模糊推論系統 15
3. CTPA分群路徑預測架構 16
3.1 CTPA分群路徑預測架構演算法 16
3.2最大期望值演算法應用於路徑分群之系統架構 23
3.3 斜率與向量差應用於路徑預測法 27
3.4 ANFIS應用於颱風終點預測之架構 28
4. 實驗設計與結果分析 31
4.1 實驗一：2009年台灣地區發佈警告之颱風路徑預測 33
4.1.1資料來源 33
4.1.2實驗起始預測點之設計 33
4.1.3蓮花颱風之實驗結果 34
4.1.4莫拉非颱風之實驗結果 37
4.1.5莫拉克颱風之實驗結果 40
4.2 實驗二：2010年台灣地區發佈警告之颱風路徑預測 43
4.2.1資料來源 43
4.2.2實驗起始預測點之設計 43
4.2.3凡那比颱風之實驗結果 44
4.2.4梅姬颱風之實驗結果 48
4.2.5莫蘭蒂颱風之實驗結果 52
4.3實驗結果分析 56
5.結論與未來研究 64
參考文獻 65
自傳 67

[1]Gaffney, S. J., Smyth, P., “Probabilistic Curve-Aligned Clustering and Predictionwith Mixture Models,”Ph.D.Dissertation, Department of Computer Science, University of California, Irvine, 2004.
[2]台大生工系水資源資訊系統研究室 : ANFIS適應性類神經網路使用說明。
[3]Aase, K. K., “A Markov model for the pricing of catastrophe insurance futures and spreads,”2009.
[4]曾龍譯，資料採礦-概念與技術，維科圖書有限公司，臺北，第 280-282頁，2003。
[5]Tan, P. N., Steinbach, M. and Kumar, V., Introduction To Data Mining , Addison Wesley, U.S.A, pp.270-271, 2006.
[6]Gaffney, S. J., Smyth, P., “Probabilistic Journal of Risk and Insurance Curve Clustering and Alignment,” Advances in Neural Information Processing Systems MIT Press68, pp.25-49, 2001.
[7]Camargo, S., Robertson, A., Gaffney, S., and Smyth, P. “Cluster analysis of Western North Pacific tropical cyclone tracks,” Extended Abstract, Proceedings of the 26th Conference on Hurricanes and Tropical Meteorology, May3-7, Miami, 10A.7, pp. 250-251,2004.
[8]Camargo, S. J.and Robertson, A. W. “Cluster Analysis of Typhoon Tracks. Part I: General Properties,” 2006.
[9]Terry, James. P. and Feng, Chen-Chieh. “On quantifying the sinuosity of typhoon tracks in the western North Pacific basin” , 2010.
[10]Han, Y. J., and Boves, L. “Trajectory clustering for automatic speech recognition,” In Proceedings of EUSIPCO-2005, September 2005.
[11]Ueda, N., and Nakano, R., “EM algorithm with split and merge operations for mixture models,” Systems and Computers in Japan, vol. 32, pp. 1–11, 2000.
[12]Robertson, A.W., Smyth, P., Camargo, S. J and Ghil,M. “Probabilistic clustering of extratropical cyclones using regression mixture models,” 2007.
[13]Frey, B. J. and Jojic , N. “Transformation-invariant clustering using the EM algorithm,” IEEE Trans. PAMI, 25(1), January 2003.
[14]Gaffney, S. J. and Smyth, P., “Curve clustering with random effects regression mixtures,” In Proc. Ninth International Workshop on Artificial Intelligence and Stats, Key West, FL, January 3–6, 2003.