空氣中PM2.5自動監測線性迴歸模式之研究-以台南地區為例

CNU IR > College of Sustainable Environment and Recreation > Dept. of Environmental Engineering and Science (including master's program) > Dissertations and Theses > Item 310902800/30041

Please use this identifier to cite or link to this item: https://ir.cnu.edu.tw/handle/310902800/30041

Title:	空氣中PM2.5自動監測線性迴歸模式之研究-以台南地區為例 Characterization of Linear Regression Models in Atmospheric Fine Particulate Matter (PM2.5) between Automatic and Manual Monitoring - A Case Study in Tainan
Authors:	林紀壯
Contributors:	環境工程與科學系蔡瀛逸王怡敦
Keywords:	多元線性迴歸細懸浮微粒平均絕對誤差百分比空氣品質監測站台南地區 multiple linear regression fine particle (PM2.5) mean absolute percentage error (MAPE) air quality monitoring station Tainan
Date:	2016
Issue Date:	2016-12-21 15:32:26 (UTC+8)
Abstract:	細懸浮微粒(PM2.5)對於人體健康影響的危害較粗懸浮微粒更為嚴重與顯著。我國行政院環境保護署於2012年5月14日公告PM2.5空氣品質標準，將24小時平均目標值訂定為35 μg/m3，年平均目標值為15 μg/m3。為保障國人身體健康，環保署於全國設立之空氣品質監測網，將懸浮微粒分為「手動監測」及「自動監測」，並於2014年5月公告細懸浮微粒自動監測數據校正原則，以手動法為標準建立兩者間關係式，用以校正自動監測數據，使兩者數據趨於一致並即時公布，提供預警功能及更正確的資訊。本研究先以102年11月至104年10月台南地區具有PM2.5手動監測的兩個空氣品質監測站（臺南站及新營站），進行PM2.5手動採樣數據及PM2.5自動監測數據之解析，再以自動空氣品質監測站之自動監測數據為自變數（包括PM2.5自動測值、PM10自動測值、溫度、濕度、風速、二氧化硫、氮氧化物、一氧化碳及臭氧），PM2.5手動測值為應變數進行「多元迴歸」(Multiple Regression)分析，最後再納入台南地區沒有PM2.5手動監測的善化站及安南站進行討論，最終數據之修正成效，以平均絕對誤差百分比(Mean Absolute Percentage Error, MAPE)為判定指標。研究結果顯示，臺南站及新營站多數PM2.5自動測值大於手動測值，而公告迴歸式（關係式）的修正雖然可使平均絕對誤差減小，但卻造成多數迴歸自動測值的低估，只有透過多元迴歸的修正，不只能下修測值還有部分測值往上修正，使迴歸自動測值更趨近於手動測值，MAPE指標可達「優?的預測能?」等級。善化站及安南站使用臺南站PM2.5手動測值做為多元線性迴歸的基礎，預測能力表現相較於臺南站較不佳，但仍屬優良的預測能力範圍。若欲達到更佳的預測能力，應以當地的PM2.5手動測值採樣分析結果做為迴歸基礎才有較佳的效果。應用多元線性迴歸方程式修正臺南站及新營站105年1月至4月監測數據，另於105年1月至3月間選定善化站執行16站次的PM2.5手動採樣監測以確認預測能力，結果顯示多元迴歸的修正效果皆較公告迴歸式（關係式）有較佳的修正效果。臺南站及善化站MAPE指標皆達「優?的預測能?」等級，新營站MAPE指標更達「高度準確的預測能?」等級，但整體而言，對於使用臺南站PM2.5手動測值採樣分析結果做為多元迴歸基礎的善化站，預測能力仍較臺南站及新營站不佳。 Fine particulate matter (PM2.5) causes much more severe health damage than coarse particles. The Environmental Protection Administration (EPA) of Taiwan published PM2.5 air quality standards on May 14, 2012, 35 ?g/m? is the target value of 24-hour average, and 15 ?g/m? for annual limit value. To ensure the protection of the citizen's health, the EPA set up air quality monitoring network, the monitoring methods for suspended particulate matter are divided into "Manual monitoring" and "Automatic monitoring", and the adjustment principle of automatic monitoring was also announced on May 2014. Referring the data from manual monitoring to adjust the automatic monitoring data, so that these two data convergence would be announced immediately, and provide early warning capabilities and accurate information.In this research, firstable, to collect the PM2.5 data from two air quality monitoring stations- Tainan station and Xinying station, and analysis the PM2.5 data which both were monitored from manual method and automatic method from November 2013 to October 2015. Then, the auto-monitoring data of automatic air quality monitoring station as independent variables (including PM2.5 automatic measured values, PM10 automatic measured values, temperature, humidity, wind speed, SO2, NOX, CO and O3), and the manual-monitoring data of PM2.5 as dependent variables in the multiple regression analysis. In the end, to include the data from other two PM2.5 monitoring stations- Shanhua station and Annan station located in Tainan which don’t have manual PM2.5 monitoring, and to discuss the effectiveness of the final correction data, then to establish the indicators of determined from the mean absolute percentage error (MAPE).The result shows that the majority of automatic measured values are more than the manual measured values no matter in Tainan station or Xinying station. Although it will reduce the mean absolute error by using linear regression equation, but it usually leads to the underestimation of the automatic measured values. With the multiple regression, not only revise downward measured values but also adjust upward measured value if needed, so that the automatic measured value would be closer to the manual measured value, the MAPE indicator will reach the level of "Good forecast".Shanhua station and Annan station adopted the manual-monitoring data of PM2.5 from Tainan station to be the multiple regression analysis, even have the good forecast, but their prediction capability was poor compared to Tainan station. In order to achieve better predictive capability, Shanhua station and Annan station should be adopted the manual-monitoring data of PM2.5 from themselves.Application of multiple linear regression equation In Tainan station and Xinying station from January to April 2016, and made 16 times PM2.5 manual monitoring in Shanhua station from January to March 2016, the results showed that the multiple regression have better predictive ability than announcement of linear regression equation.The MAPE value of Tainan station and Shanhua stationcan reach the level of "Good forecast", and Xinying stationcan reach the level of "highly accurate forecast". But overall, Shanhua station adopted the manual-monitoring data of PM2.5 from Tainan station to be the multipleregression, it’s prediction capability was poor compared to Tainan station and Xingying station.
Relation:	學年度：104，119頁
Appears in Collections:	[Dept. of Environmental Engineering and Science (including master's program)] Dissertations and Theses

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	1193	View/Open

Loading...