應用類神經網路預測雲林地區光化學污染物之研究

Predicting Photochemical Pollutants in Yunlin Area Using Propagation Neural Network

高敬棠(Ching-Tang Kao)

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(2011年)

摘要

本研究應用倒傳遞類神經網路（Back-Propagation Neural Network, BNN），建立雲林地區空氣品質之預測模式，變數因子分別為臭氧(Ozone, O3)、一氧化碳(Carbon monoxide, CO)、氮氧化物(Nitrogen oxides, NOx)、一氧化氮(Nitric oxide, NO)、二氧化氮(Nitrogen dioxide, NO2)、二氧化硫(Sulfur dioxide, SO2)、Temp(Temperature, Temp)，以2010年1月~11月為網路輸入参數建立最適化網路，對2010年12月之O3及NOx作預測，研究結果顯示O3預測部份以台西6個影響參數對應1個輸出參數的平均絕對百分比誤差值（Mean Absolute Percentage Error, MAPE）18.32%為最高，其次為崙背3個影響參數對應1 個輸出參數MAPE值20.97%，又其次為斗六2個影響參數對應1 個輸出參數MAPE值25.55%。NOx預測部份以斗六6個影響參數對應1 個輸出參數MAPE值17.74%為最高，其次為崙背4個影響參數對應1 個輸出參數MAPE值20.46，又其次為台西7個影響參數對應1個輸出參數MAPE值24.56，O3及NOx預測之結果良好，表示BNN模型對於預測濃度及變動趨勢皆可掌握。

關鍵字

倒傳遞類神經網路 ； 平均絕對百分比誤差值 ； 氮氧化物 ； 臭氧

並列摘要

This study employed Back-Propagation Neural Network (BNN) method to establish an air quality prediction model of Yunlin area. Variable factors used Ozone, Carbon monoxide, Nitrogen oxides, Nitric oxide, Nitrogen dioxide, Sulfur dioxide, Temp. We input data between January and November, 2010 as parameters to establish an optimizing network to predict the air quality of O3 and NOx on December, 2010. The research shows that the best mean absolute percentage error (MAPE) 18.32% by using six input parameters to compare one output parameter in O3 of west sub-bureau. The second MAPE 20.97% by using three input parameters to compare one output parameter in O3 of Lunbei. Last MAPE 25.55% by using tow input parameters to compare one output parameter in O3 of Douliou. In predict part, the best result will be MAPE 17.74% by using six input parameters to compare one output parameter in NOx of Douliou. The next will be MAPE 20.46% by using four input parameters to compare one output parameter in NOx of Lunbei. The last one will be MAPE 24.56% by using seven input parameters to compare one output parameter in NOx of west sub-bureau. According to training and prediction of O3 and NOx figure, the simulation result is better. Therefore, the mode makes a conclusion that predicts effect perfects.

並列關鍵字

BNN ； MAPE ； Nitrogen oxides ； Ozone

參考文獻

15. Acuna G., H. Jorquera, R. Perez, ”Neural network model for maximum ozone concentration prediction., ”Lecture Notes in Computer Science, No.1112, pp.263-268, (1996).

16. Comrie A.C., ”Comparing neural networks and regression models for ozone forecasting, ”Journal of Air and Waste Management, Vol.47, pp.653-663, (1997).

17. Chang S.C., Pai T.Y., Ho H.H., Leu H.G. and Shieh Y.R. Evaluating Taiwan’s air quality variation trends using grey system theory. Journal of the Chinese Institute of Engineers, 30(2), 361-367 (2007).

18. Pai T.Y., Ouyang C.F., Su J.L. and Leu H.G. Modelling the steady-state effluent characteristics of the TNCU process with ASM2d under varied SRT conditions. Journal of the Chinese Institute of Environmental Engineering, 10(1), 35-42 (2000a).

20. Pai T.Y., Ouyang C.F., Su J.L. and Leu H.G. Modeling the stable effluent qualities of the A2O process with Activated Sludge Model 2d under different return supernatant. Journal of the Chinese Institute of Engineers, 24(1), 75-84 (2001a).