短期空氣品質應變措施於多種天氣型態之有效性分析與效益評估—以北部空品區為例

Abstract

高濃度空氣污染事件可能增加民眾的健康風險、對健康產生立即性的危害，因此許多國家開始實施空氣品質管理的短期應變計畫；我國政府也在民國106年修正了<空氣品質嚴重惡化緊急防制辦法>，希望以短期空氣品質管理應變措施，減緩高濃度事件日的時間與濃度值；然而目前關於短期應變措施的研究較少，且機制模型存在短期的預測限制，無法有效得知短期作為對於實際濃度改變情形。 本研究目的在於評估短期作為成效之研究方法上的不足，以北部空品區為研究地點，選定在三種天氣型態下的降載策略及一交通管制策略為案例，其中降載作為中減量的目標污染物PM10與PM2.5濃度、交通作為中目標污染物NO2、O3濃度；以深度學習的長短期記憶模型(Long Short-Term Memory, LSTM)預測無短期作為下的空氣品質惡化濃度，再利用統計檢定方法分析應變作為前後的濃度區間範圍及濃度減量的可能，並以空氣污染物傳輸模式模擬電力業降載時的影響範圍與效益，作為空間中的輔助資訊。結果顯示，預測模型在後一小時、後三小時中表現極佳；其中北部空品區禁車的短期應變作為確實使得當日的NO2與O3濃度觀測值下降，對於降載案例中的PM10及PM2.5濃度而言，實施短期應變作並沒有辦法使得濃度明顯降低，但以有效性濃度差值解釋，高壓迴流與鋒前暖區兩個天氣型態下，降載作為可使得PM10及PM2.5濃度下降的機率顯著增加；對於高壓迴流的天氣型態而言，降載影響區域較為局部，而鋒前暖區的天氣型態在全區中有較廣泛的削減百分比，高壓推擠型則沒有明顯的降載成效。 本研究創新利用深度學習的方法對真實減量濃度作出解釋，輔助以空氣污染物傳輸模型探討影響範圍與削減受益地區，以濃度數值與空間分析兩個面向，將既有方法結合新興數據科學的分析方式，對於策略擬定及短期應變作為效益分析提供參考依據。

High air pollutant incidents increase people's health risks definitely, and the possibility of immediate harm to health may also increase. At present, more and more countries are beginning to implement short-term plans for air quality management. Our government also revised the 'Emergency Prevention Measures for Severe Deterioration of Air Quality' in 2017, by utilizing short-term measures as a mean to reduce the time and concentration of high-concentration event days; however, few studies on short-term contingency measures were found; assessments on short-term measures also faced limitation due to high variation of small-scale emissions and meteorological conditions in a timely manner. The aim of this study is to improve the deficiencies in the assessment of short-term contingency measures. With the northern air quality zone of Taiwan selected as study area, this study analyzed the validity analysis and benefit evaluation of the load reduction of power plant strategy within three weather patterns for target reduction of PM10 and PM2.5 concentration, and a traffic strategy for target reduction of NO2 and O3 concentration. First, the deep learning prediction model simulates the air quality deterioration concentration under the baseline condition (no pollutant reduction action), and the statistical analysis method is used to test the possibility and effectiveness of target pollutant reduction strangies; then, using air the pollutant transport mode simulates the impact range of the power industry and the efficiency in different regions during load shedding. The results showed that the model prediction ability performed extremely well within one hour and three hours after the prediction. The traffic case showed that the zero-car event of short-term plans in the northern air quality zone is indeed the effect of reducing the NO2 and O3 concentration observations on the day, yet the load reduction of power plant strategy is not implemented as the apparent concentration for the PM10 and PM2.5 concentrations. The concentration range was used to explain the difference between the two weather patterns: High Pressure Peripheral Circulation (HPPC) and Warm Area Ahead of a Front (WAF), and the implementation of power plant emission reductions in both weather patterns tended to reduce the concentration of PM10 and PM2.5. HPPC affected areas were more local and the WAF had a relatively average reduction percentage in the whole area, while the High Pressure System Pushing (HPP) type had no obvious load reduction effect. This research uses the deep learning method to explain the real concentration reduction, and assists the air pollutant transport model to explore the scope of impact and reduce the benefit areas. Combining the existing methods and the analysis methods of emerging data sciences, the two aspects of concentration and spatial analysis provide reference for strategy formulation and short-term air quality contingency measures as benefit analysis.