臭氧是台灣地區光化學污染的主要產物，由前驅物質(主要為NOx和VOC)經大氣化學反應所形成的二次污染物質，其生成是一複雜的非線性反應，許多氣象因子如地表溫度、日照強度、雲量、風速、風向、相對溼度等均扮演重要的角色，而且隨著季節變化，不同地區的天氣型態所造成的臭氧事件日也有所不同，所以本研究期望藉由類神經網路來建立臭氧預測模式，以作為臭氧濃度預報系統。
研究中以傳統統計預報及完美預報來進行次日最大小時臭氧濃度及八小時平均臭氧濃度的預報工作，並將預報結果和實測值比較，結果顯現出以完美預報所得的結果比傳統統計預報為佳，此外，以完美預報所預測的八小時平均臭氧濃度與實測值之相關係數均能達到高度相關，然而所有預測均出現高估低濃度值、低估高濃度值的現象。

The ground-level ozone produced by photochemical air pollution is a serious environmental problem in Taiwan. It is a secondary pollutant generated by precursors (mainly NOx and VOC) through serial complicated reactions with other chemical species in the atmosphere. The forming of ozone is a complicated and non-linear reaction. Other than the chemical reaction, the weather condition, such as surface temperature, sunlight strength, cloudiness, wind speed, wind direction and humidity, also play an important role in the photochemistry pollution. This study focuses on building a model that predicts the next day’s maximun ozone concentration with artificial neural network.
In the study, two forecasts, i.e., classical statistical and perfect prog forecast, were used to predict the next day’s maximum hourly and eight hour average ozone concentration prediction. The study found that the results obtained from perfect prog forecast are more accurate than that obtained from classical statistical forecast. The results of all forecasts are very reasonable. High correlations between the predictions of perfect prog forecast and measured 8h ozone concentrations are noted.  However, all forecasts tend to overpredict in the low concentrations and underpredicted in the high concentrations.

目錄
中文摘要	I
英文摘要	II
目錄	III
表目錄	V
圖目錄	VI
第一章 前言	1
1.1 研究動機	1
1.2 文獻回顧	4
  1.2.1 臭氧形成機制	4
  1.2.2 過去常用的預報方法	5
  1.2.3類神經網路在空氣品質預報上的應用	6
  1.2.4氣象上統計預報的方法	8
1.3 研究目的	10
第二章 類神經網路的介紹	11
2.1 何謂類神經網路	11
2.2 類神經網路的優缺點	12
2.3 類神經網路的應用	14
2.4 類神經網路的分類	15
  2.4.1 依學習方法分類	15
  2.4.2依網路架構分類	16
2.5 類神經網路的學習演算法	18
第三章 研究方法	21
3.1測站選取	21
3.2資料選取	25
3.3倒傳遞類神經網路	29
  3.3.1原理公式	29
  3.3.2參數設定	33
3.4輸入的變數	35
第四章 結果討論	37
4.1評估指標	37
4.2分析之模式	40
4.3預測最大小時臭氧濃度結果分析	41
  4.3.1 完美預報	41
  4.3.2 傳統統計預報	44
  4.3.3 完美預報與傳統統計預報的比較	45
4.4預測八小時平均臭氧濃度結果分析	52
  4.4.1 完美預報	52
  4.4.2 傳統統計預報	53
  4.4.3 完美預報與傳統統計預報的比較	55
第五章 結論與建議	62
參考文獻	64
附錄A  最大小時臭氧濃度關係圖(完美預報)	68
附錄B  最大小時臭氧濃度關係圖(傳統統計預報)	71
附錄C  八小時平均臭氧濃度關係圖(完美預報)	74
附錄D  八小時平均臭氧濃度關係圖(傳統統計預報)	77
表目錄
表1.1 台灣地區空氣品質標準	2
表1.2 各國空氣品質標準	3
表2.1 各領域之類神經網路應用	14
表2.2 學習演算法	18
表3.1 空氣品質監測站之分區	24
表3.2 輸入參數	28
表3.3 參數設定資料	34
表4.1 第n日氣象條件預測第n日最大小時臭氧濃度之評估指標	46
表4.2 第n-1日氣象條件預測第n日最大小時臭氧濃度之評估指標	49
表4.3 第n日氣象條件預測第n日最大八小時平均臭氧濃度之評估指標	56
表4.4 第n-1日氣象條件預測第n日最大八小時平均臭氧濃度之評估指標	59
圖目錄
圖2.1 監督式學習法示意圖	16
圖2.2 非監督式學習法示意圖	16
圖2.3 前饋式網路架構示意圖	17
圖2.4 回饋式網路架構示意圖	17
圖3.1-1 空氣品質監測站分布圖	22
圖3.1-2 空氣品質監測站北部分布圖	23
圖3.1-3 空氣品質監測站南部分布圖	23
圖3.2 遞類神經網路架構圖	33
圖3.3 BP演算法傳播示意圖	33
圖3.4 完美預報示意圖	36
圖3.5 傳統統計預報示意圖	36
圖4.1 位於北部沿海地區之測站	42
圖4.2-1 台灣內陸地區之測站	42
圖4.2-2 台灣內陸地區之測站	42

Acuna G., Jorquera H., Perez R., 1996, Neural network model for maximum ozone concentration prediction., Lecture Notes in Computer Science, 1112, 263-268.
Abdul-Wahab, S.A., Al-Alawi, S.M., 2002, Assessment and prediction of tropospheric ozone concentration levels using artificial neural networks., Environmental Modelling & Software, 17, 219–228.
Comrie, A. C., 1997, Comparing neural networks and regression models for ozone forecasting., Journal of Air and Waste Management, 47, 653-663.
Cassmassi J.C., 1987, Development of an objective ozone forecast model for the South Coast Air Basin. Presented at the Air Pollution Control Association 80th Annual Meeting, New York, NY, June 21-26.
Dye T.S., Ray S.E., Lindsey C.G., Arthur M., and Chinkin L.R., 1996, Summary of ozone forecasting and equation development for the air districts of Sacramento, Yolo-Solano, and Placer. Vol. I: ozone forecasting. Vol. II: equation development. Final report prepared for Sacramento Metropolitan Air Quality Management District, Sacramento, CA by Sonoma Technology, Inc., Santa Rosa, CA, STI-996210-1701-FR, December.
EPA., 2003, Guidelines for Developing an Air Quality (Ozone and PM2.5) Forecasting Program., U.S. Environmental Protection Agency, Office of Air Quality Planning and Standards, Research Triangle Park, NC, EPA-456/R-03-002
Hubbard M.C. and Cobourn W.G., 1997, Development of a regression model to forecast groundlevel ozone concentration in Louisville., KY. Atmos. Environ. (submitted).
Horie Y., 1988, Air Quality Management Plan 1988 Revision, Appendix V-P: Ozone episode representativeness study for the South Coast Air Basin. Report prepared for the South Coast Air Quality Management District, El Monte, CA by Valley Research Corporation, Van Nuys, CA, VRC Project Number 057, March.
Kasten, F., Czeplak, G., 1980. Solar and terrestial radiation dependent on the amount and type of cloud. Solar Energy, 24, 177-189.
National Research Council, 1991 Rethinking the Ozone Problem in Urban and Regional Air Pollution, National Academy of Sciences/National Research Council, National Academy Press, Washington, DC.
Prybutok, V. R., J. Yi, and D. Mitchell, 2000, Comparison of Neural Network Models with ARIMA and Regression for Prediction of Houston’s Daily Maximum Ozone Concentrations., European Journal of Operational Research, Vol. 122, 31-40.
Robinson, R., 1991, Neural networks offer an alternative to traditional regression., Geobyte February, 14-19.
Ryan W.F., 1994, Forecasting severe ozone episodes in the Baltimore metropolitan area. Atmos., Environ. 29, 2387-2398.
Seinfeld, J.H., and S.N. Pandis. 1998. Atmospheric Chemistry and Physics, From Air Pollution to Climate Change. John Wiley and Sons, Inc., NY.
Sohn, Sang Hyum, Oh, Sea Cheon, Jo, Byung Wan, and Yeo, Yeong-Koo, 2000, Prediction of Ozone Formation Based on Neural Network., Journal of Environmental Engineering, 688-696,.
Stoeckenius T., 1990, Adjustment of ozone trends for meteorological variation., Presented at the Air and Waste Management Association's Specialty Conference, Tropospheric Ozone and the Environment, Los Angeles, CA, March 19-22.
Spellman, G., 1998, An Application of Artificial Neural Networks to the Prediction of Surface Ozone Concentrations in United Kingdom., Applied Geography, Vol. 19, 123-136.
Wilks, D.S., 1995, Statistical Methods in the Atmospheric Sciences, Academic Press.
Yi, J. and Prybutok, R., 1996, A neural network model forecasting for prediction of daily maximum ozone concentration in an industrialised urban area., Environmental Pollution, 92(3), 349-357.
黃宗仁，利用類神經網路預測台中都會區臭氧趨勢之研究，碩士論文，國立中興大學環境工程研究所，2001。
黃尚雄，應用類神經網路預測空氣品質之研究，碩士論文，國立交通大學，1995。
曹勝雄，曾國雄，江勁毅，傳統計量迴歸、模糊迴歸、GMDH、類神經網路四種方法在預測應用上之比較-以國人赴港旅客需求之預測為例，中國統計學報第34 卷第2 期，pp.132-161，1996。
陳順宇、鄭碧娥，統計學，華泰書局，三版，台北，第14-1~15-18頁，1998。
林茂文，時間數列分析與預測，華泰書局，台北，第11-185頁，1992。
葉怡成，類神經網路模式應用與實作第七版，2000。
楊肇政、鄭阿全，污染防制，高立圖書有限公司，台北，第84-87 頁，1990。
張斐章、張麗秋，類神經網路，東華書局，台北，2005。
經濟部水利署，智慧型水文防洪系統建置，2004。
行政院環保署網站，http://www.epa.gov.tw。
中央氣象局網站，http://www.cwb.gov.tw。