結合光學量測技術與多變量統計方法追蹤調查工業區異味污染源

Abstract

近年來環保意識抬頭，民眾對於工廠排放的有害或異味空氣污染物的關切與日俱增，異味陳情件數也因此大幅增加。由於紅外光遙測技術兼具長時間連續即時且大範圍的量測，可同時監測多種有機和無機污染(含有害及異味空氣)成分等技術特性，當結合同步觀測的風速風向等氣象資料時，可以迅速標定污染的成分和來源方向，因此近年來也普遍地應用於環境異味事件的調查工作。然而，由於該技術可同時監測的污染物種類繁多，產生的資料量龐大，傳統上以單一成分或兩兩之間相關性進行資料的解析，容易遺漏污染源排放特徵等重要資訊。本研究透過主成分分析(principal component analysis, PCA)、因素分析(factor analysis, FA)、群集分析(cluster analysis, Cluster)及對應分析(correspondence analysis, CA)等多變量統計分析技巧，以獲得具有污染源排放特徵之因素群組，搭配風向及風速等氣象資料，歸納污染源的類型及污染的原因。 本論文以(石化)工業區與周邊的社區為研究對象，藉由於異味熱區設計光學佈線，先找出可能的異味來源方向，再進行上風處來源工廠之排放管道或製程區的調查，比對環境中異味成分的因素群組和工廠排放/逸散氣體成分的因素群組之間的關係，進一步釐清異味污染成份及造成異味排放的可能原因。第一個主題從工業區周界外異味敏感區為起點，透過光學遙測儀器的多重佈線、氣象資料及主成分分析的應用，循序漸進地回溯與限縮(backtracking and refining)污染源的範圍，再運用Closed-cell-FTIR (以下簡稱CC-FTIR)進行污染源排放組成的連續量測。在為期3個月的監測共蒐集20,401筆監測資料，發現環境中二甲基甲醯胺（DMF）、丁酮，甲苯和乙酸乙酯的平均值濃度偏高；其中DMF的濃度已超過固定污染源排放管制標準。利用PCA確定受體端有三種排放源分別為PU塗佈、化學包裝和平版印刷行業；利用CC-FTIR確認DMF的來源與PU塗佈廠有關。第二個主題透過石化製程區矩陣式OP-FTIR多重佈線、PCA、CA與氣象資料的整合應用，探討石化廠不同製程區的污染特性以歸納主要逸散源之所在，輔助傳統點監測無法全面掌握製程區洩漏源的缺憾。在石化製程區內佈線17條矩陣式OP-FTIR測線，鑑定出苯乙烯、丁二烯、環己烷、氨、乙烯、丙烯和甲醇。 PCA歸納出工廠的製程區至少有三種污染來源類別；CA提供每個排放源大概的位置。本研究通過結合PCA和CA的OP-FTIR矩陣路徑方法來定位複雜石化製程區的逸散性污染源。第三個主題是從工業區內的異味敏感點為起點，透過單一OP-FTIR佈線、FA及氣象資料與異味記錄，快速標定主要異味的來源方向，再以定點式的光學監測儀器，進行排放管道連續監測數據的蒐集，結合FA及Cluster等多變量統計分析技巧，比對環境中異味成分的因素群組和工廠排放氣體成分的因素群組之間的關係，利用Cluster的樹狀圖分析成分間分離和融合的關係，構建一個可以解析管道氣來源結構的資訊，歸納造成異味污染的可能原因。為期10天的監測共蒐集2,911筆連續的濃度及氣象資料，發現環境中汽油，間二甲苯，二氧化氮，乙酸正丁酯，甲苯和PGMEA的濃度高於嗅覺閾值，顯示這些化合物可能是造成異味的可能原因。FA顯示異味成分的主要用途為表面塗料或油漆中的有機溶劑。CC-FTIR在附近的兩家表面塗裝工廠進行管道監測，共蒐集4,378筆連續的排放管道監測資料，顯示只有一家公司的排放管道呈現與受體相同的因素組合。第四個主題是針對焚化型的排放管道，以密閉式光學量測技術進行長時間連續的監測，配合FA以探討影響污染排放的可能原因，並透過因素分數的計算，解析不同因素群組的時序變化特徵，解析造成因不完全燃燒而產生高異味排放的可能原因。

With the rising awareness of environmental protection and the increase in the proportion of respiratory diseases in recent years, the public’s concern regarding harmful or odorous pollutants emitted by factories has increased. Consequently, the number of complaints about odor nuisances has drastically increased. Fourier Transform Infrared (FTIR) spectrometry provides the long-term continuous and large-scale monitoring of pollution and simultaneously monitoring of various organic and inorganic pollution (including harmful and odorous air) components. When meteorological data (wind direction and wind speed) are incorporated, FTIR can determine the composition and source direction of the pollutants. Therefore, FTIR has been widely applied in the investigation of environmental odor nuisance complaints in recent years. Conventional data analysis methods based on single components or pairwise correlation may miss crucial information, especially regarding the emission characteristics of pollution sources. In the present study, multivariate statistical analysis techniques such as principal component analysis (PCA), factor analysis (FA), cluster analysis (Cluster), and correspondence analysis (CA) were employed. Dimensionality reduction of data through the multivariate statistical model can generate factor groups with the emission characteristics of pollution sources. Given that the identification of the emission sources responsible for industrial VOCs odors remains a great challenge, this research demonstrated that the odor emission sources could be characterized and located from a systematic odor investigation framework. This research developed a series of investigation frameworks to locate both stationary and fugitive emission sources using a dual-optical sensing system, matrix-path approach, and multivariate statistical modeling. The backtracking & confining methods was developed for characterizing and locating complex sources in an industrial setting. During the 3-month monitoring period, a total of 20,401 spectrum data were collected. Dimethylformamide (DMF), butanone, toluene, and ethyl acetate displayed high concentrations in the environment. The concentration of DMF, in particular, exceeded the fence line standards for stationary sources. The results of PCA and Closed-cell FTIR (CC-FTIR) revealed that the source of DMF was associated with PU coating. The large-scale search method was developed using a matrix-path approach to confine a large fugitive emission area into several smaller sub-regions to make the usage of traditional point sampling methods more effective and efficient. A total of 17 matrix OP-FTIR lines were deployed at the petrochemical manufacturing areas, in which styrene, butadiene, cyclohexane, ammonia, ethylene, propylene, and methanol were identified. PCA results indicated that the manufacturing areas of the plants contained at least three types of pollution sources, and CA results provided the approximate location of each source. The source–receptor profiling method was developed for identifying the major emission sources and determining emission profiles of each stack source. FA and cluster analysis were combined to determine the relationship between the factor groups of odor components in the environment and those of stack exhaust gas components. During a 10-day monitoring period, a total of 2,911 continuous concentration and meteorological data were collected, which suggested that the concentrations of gasoline, m-xylene, nitrogen dioxide, butyl acetate, toluene, and PGMEA exceeded the odor thresholds. This indicated that the aforementioned compounds might have been the causes of odor. FA demonstrated that odorous components were mainly used in organic solvents in surface coatings or paints. The stack monitoring was conducted at two nearby surface coating plants using a CC-FTIR, and a total of 4,378 continuous emission monitoring data were collected. Only the emission of one plant exhibited the same odorous substance profile found by the OP-FTIR receptor path. The major odor emission source was therefore identified. The source apportionment method for combustion stacks was developed for the apportionment of harmful/odor pollutants from different combustion stack emissions. The field measurements involved the long-term continuous monitoring of the exhaust gas from an incinerator using a CC-FTIR with the application of FA to determine the possible causes of odor emission. The time-series characteristics of different factor groups were extracted by calculating the factor scores to reveal the possible causes of odor emissions. Field studies demonstrated the feasibility of the proposed methods.