第一章绪论 - 康宁大学.doc

應用類神經網路預測台南地區氮氧化物(NOx) 吳春生 摘 要 依據環保署訂定的空氣污染監測指標為：懸浮粒子(PM10)、二氧化硫(SO2)、一氧化碳(CO)、氮氧化物(NOx)、臭氧(O3)、鉛(Pb)等六項，其中以氮氧化物為臭氧生成的前驅物，並會增加罹患呼吸系統疾病的機率，以及產生酸雨。假若可以預測氮氧化物生成趨勢，就可以做為空氣品質預警系統之參考。 因此，本研究利用類神經網路(ANN)在非線性資料處理上較為優勢的特點與傳統時間序列法分別對台南地區的NOx濃度進行預測，並加以分析比較。在整體預測值上，類神經網路的預測R值（相關係數）已達到0.8以上，而時間序列法的預測R值則會隨著時間的增加而降低，所以在預測能力上類神經網路會比時間序列法來的優秀。 而後續研究可結合現有環保署或環保局的測站數據，進行監測網的多目標規劃分析，以評估空氣品質監測效益，和深入分析每小時空氣監測值，瞭解一天當中氮氧化物的濃度變化，或是分季節進行研究，以因應季節不同而調整空氣品質的管制策略。 關鍵字：空氣品質、氮氧化物、類神經網路、時間序列 The Application of Artificial Neural Network for Predicting NOx of Tainan Area Yu-Chin Ling Chun-Sheng Wu Abstract The indicators of monitoring air pollution created by Environmental Protection Administration are: PM10, SO2, CO, NOx , O3 and Pb. Among these indicators, NOx is the precursor of forming O3 and it will increase the possibility of getting respiration system diseases and produce the acid rain. If it is possible to predict its forming trend, the trend can be used to be a reference of the air quality. As a result, the research employs methods of the Artificial Neural Network(ANN) which is superior in dealing with the non-linear data and the traditional Time Sequence to predict and analyze the consistency of NOx in Tainan area. In the overall predictive values, the R value(correlation) of the ANN is above 0.8 and, on the contrary, the R value of the Time Sequence will decrease as the time goes by. So the predictive ability of ANN is better than that of the Time Sequence. In the future, the monitoring data of Environmental Protection Administration or Environmental Protection Bureau can be applied in successive research to evaluate the advantages of monitoring the air quality and to further analyze data of air monitoring in every hour to understand the variation of NOx consistency. Or the further research can be done in different seasons and the controlling strategies is able to be adjusted by different seasons. Keywords: the air quality, NOx, Artificial