以倒傳遞類神經網路及多元線性迴歸模擬建築工地對台中縣粒狀污染物之影響

PM10 Impact Simulation of Construction Sites on Taichung County Using Back-Propagation Neural Network (BNN) Method and Multiple Linear Regression (MLR) Method

阮成隆(Cheng-Lung Juan)

指導教授 ： 白子易

朝陽科技大學/理工學院/環境工程與管理系/碩士(2010年)

摘要

本研究應用倒傳遞類神經網路（Back-Propagation Neural Network, BNN）與多元線性迴歸（Multiple Linear Regression, MLR）建立台中地區空氣品質之預測模式，BNN及MLR變數因子分別為以懸浮微粒（PM10）、懸浮微粒（PM2.5）、建築（房屋）工程（RC）、 建築（房屋）工程（SRC）、建築（房屋）工程（拆除）、隧道工程等條件，以2008年1月~11月為網路輸入參數建立最適化網路，對2008年12月之PM10作預測，並與MLR作比較。本研究利用BNN及MLR模擬預測台中地區PM10模擬結果，BNN及MLR模型對於PM10之濃度及變動趨勢皆可掌握，其中以MLR之預測效果較佳；BNN以2V1訓練及預測所得較佳，其影響PM10自變數包括PM2.5及PM10。MLR訓練以5V1訓練所得較佳，其影響PM10自變數較深在大里及豐原包括PM2.5、PM10、建築（房屋）工程（RC）、建築（房屋）工程（SRC）、隧道工程，在沙鹿為PM2.5、PM10、建築（房屋）工程（SRC）、建築（房屋）工程（拆除）、隧道工程。MLR預測結果大里、豐原及沙鹿以3V1所得較佳，影響PM10自變數包括PM2.5、PM10、隧道工程。由97年營建工程告發之工程類型中，建築（房屋）工程（RC）約佔13％、建築（房屋）工程（SRC）約佔28％，顯示與本研究上述影響PM10自變數相同，故若能對建築（房屋）工程（RC）、建築（房屋）工程（SRC）、建築（房屋）工程（拆除）、隧道工程等營建工程類型，對其加強管制應能較有效管制PM10之逸散污染。

關鍵字

台中縣 ； 懸浮微粒（PM10） ； 模糊神經網路 ； 多元線性迴歸

並列摘要

This study employed Back-Propagation Neural Network (BNN) method and Multiple Linear Regression (MLR) method to establish an air quality prediction model of Taichung area. Variable factors of BNN and MLR used PM10, PM2.5, building constructions using Reinforced Concrete(RC), building constructions using Steel Reinforced Concrete(SRC), building dismantling and tunnel constructions as input parameters. We input data between January and November, 2008 as parameters to establish an optimizing network to predict the air quality of December, 2008, and compared to MLR. The study used the simulation results of BNN and MLR models to predict the air quality of PM10 in Taichung area. BNN and MLR models both gave good simulation results toward concentrations and variations of PM10, and MLR had a better simulation result compared to BNN. BNN had better training and prediction results using 2V1 training and prediction, and affect the independent variables such as PM10 and PM2.5. MLR had a better training result using 5V1 training and prediction We found that, it affected independent variables including PM10 and PM2.5, building constructions using Reinforced Concrete(RC), building constructions using Steel Reinforced Concrete(SRC) and tunnel constructions in DaLi and FongYuan; and in SaLu, it affected independent variables including PM10 and PM2.5, building constructions using Reinforced Concrete(RC), b building dismantling and tunnel constructions. MLR had better prediction result using 3V1 and affect the independent variables such as PM10 and PM2.5. We also analyzed the construction sites which fine by the EPB in 2008 and found that 13% of them were the type of Reinforced Concrete(RC) and 28% of them were the type of Steel Reinforced Concrete(SRC). It showed that these two types match our study result. If we can control the fugitive sources of building constructions using Reinforced Concrete(RC), building constructions using Steel Reinforced Concrete(SRC), building dismantling and tunnel constructions, we will be able to reduce the emission of fugitive PM10.

並列關鍵字

Multi-regression ； Fuzzy neural network ； PM10 ； Taichung County

參考文獻

衛宇科技股份有限公司，「97年臺中縣營建工程污染管制計畫」，臺中縣環境保護局委辦計畫，台中（2008）。

胡偉興，「大型營建裸露地PM10與PM2.5排放特性之研究」，碩士論文，國立台北科技大學，台北（2006）。

Evans J.S. and D.W. Cooper, “An inventory of particulate emissions from open sources,” JAPCA,30(12):1298-1303-89-(1980).

Brook J.R., T.F. Dann, and R.T. Rurnett, “The relationship among TSP, PM10, PM2.5, and inorganic constituents of atmospheric particulate matter at multiple canadian locations.,” Journal of Air ＆Waste Management. Assoc., Vol.47, (1997).

Sohn S.H., S.C. Oh, B.W. Jo, and Y.K. Yeo, “Prediction of ozone formation based on neural network, ” Journal of Environmental Engineering, 688-696, (2000).