細懸浮微粒是粒徑小於等於2.5 μm/m3 的懸浮微粒，易深入到人體肺部，並藉由肺泡中的微血管進入到血液中，而對人體造成危害。在流行病學研究中，也發現細懸浮微粒對人體肺部疾病的發生率和死亡率有相當的影響。行政院環境保護署設置之空氣品質監測網中，有關細懸浮微粒資料之監測，至今已有長達七年的數據。本研究針對此數據，檢視細懸浮微粒監測資料之品質，並進一步建立其濃度變化趨勢之時間序列模型。以了解不同類型測站，及南北地區細懸浮微粒監測資料之可靠度，不同地區與測站間濃度變化之差異。在監測資料可靠度方面，針對缺值與儀器異常之間隔時間，利用存活分析之方法，建立對應之存活函數模型，並用對數排序檢定(Log-Rank test) 南北地區和測站類型間存活函數模型差異。在同時考慮測站類型、測站地區、故障所發生的季節、每日時段，以及故障修復的時間等因子下，建立Cox 危險比例模型(Cox proportional hazards model)，並探討各因子對監測資料可靠度的影響。在確認檢驗資料的可靠度後，進一步對細懸浮微粒濃度的變化，建立時間序列模型，預估未來的趨勢，以提供環保署做為細懸浮微粒監測數據品質評估，及未來對細懸浮微粒管制策略之參考。

Fine suspended particulate (PM2.5) is the suspended particulate where particle size is less than 2.5 μm/m3. It goes deep into the lungs easily and through alveolar capillaries enters the blood circulation, then harms the human body. The epidemiology study has discussed that fine suspended particulate has influence to incidence and death rate of lungs disease. The EPA in Taiwan has been monitoring the PM2.5 for sometime has setup air quality monitoring network. It has recorded at least seven years in all stations. This study is aimed at invest fating trend of PM2.5 through analysing the monitoring data, to view the reliability and modeling the trend in last few years. In order to find out the effect of type and location of sites to the reliability and the trend of the data observed. In this work we use the survival analysis to study the reliability of the data through intervals between missing values and unusual instrument values, then modeling the corresponding survival model with the interval values as the response time. The Log-Rank test has been used to test the effect of type and location of sites, separately. Considering the factors with type and location of sites, seasonal patterns, hourly patterns and repaired time, we further build corresponding the Cox proportional hazards model to identify the effect of factors. After checking the reliability of data, build the time series model of the PM2.5 data. Later the fitted time series model is used to forecast the future trend, which may be useful to the EPA, as reference to the environment control quality of PM2.5.

論文審定書i
誌謝ii
摘要iii
Abstract iv
1 前言1
2 資料敘述1
2.1 資料來源. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2 相關名詞. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2.1 測站類型. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.2.2 無效值. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2.3 資料選取與處理. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3.1 測站選取. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3.2 儀器壽命. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3.3 儀器存活壽命. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.3.4 白天日加權平均. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 研究方法5
3.1 混合常態分佈. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.2 Cox 比例危險迴歸模型. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3.3 主成分分析. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.4 時間序列模型. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
4 實證分析9
4.1 儀器壽命分析. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1.1 常見分佈配適結果. . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
4.1.2 K-means 分群與混合常態分佈配適結果. . . . . . . . . . . . . . . . . 11
4.2 儀器存活壽命分析. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
4.2.1 Kaplan-Meier 估計式與差異性比較. . . . . . . . . . . . . . . . . . . 13
v
4.2.2 Cox 比例危險迴歸模型. . . . . . . . . . . . . . . . . . . . . . . . . 17
4.3 濃度模型與預測. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3.1 主成份分析. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
4.3.2 時間序列模型配適. . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
5 評估與探討23
參考文獻24
附錄一25
附錄二33

[1] Benaglia, T., Chauveau, D., Hunter, D.R. and Young,D. (2009). mixtools: An R Pack-
age for Analyzing Finite MixtureModels. Journal of Statistical Software, 32(6), 1-29.
[2] Fernandez, C. and Steel, M.F. (1998). On Bayesian Modeling of Fat Tails and Skewness.
Journal of the American Statistical Association, 93(441), 359-371.
[3] Friedman, J., Hastie, T. and Tibshirani, R. (2009). The Elements of Statistical: Data
Mining, Inference, and Prediction, 2nd. Springer, New York.
[4] Kleinbaum, D.G. and Klein, K. (2005). Survival Analysis: A Self-Learning Text, 2nd.
Springer, New York.
[5] Rencher, A.C. (2002). Methods of Multivariate Analysis, 2nd. Wiley, New York.
[6] Tsay, R.S. (2005). Analysis of Financial Time Series, 2nd. Wiley, New York.
[7] United States Environmental Protection Agency-Particulate Matter
http://www.epa.gov/pm/health.html
[8] 行政院環境保護署-空氣品質監測網
http://taqm.epa.gov.tw/taqm/zh-tw/default.aspx