以政策檢驗與機器學習建模探討焚化爐減排因子

Exploring Factors for Emission Reduction in Incinerators through Policy Examination and Machine Learning Model Establishment

張千晴(Chien-Ching Chang)

指導教授 ： 闕蓓德

國立臺灣大學/工學院/環境工程學研究所/碩士(2024年)

https://doi.org/10.6342/NTU202403045

摘要

垃圾焚化為台灣主要的廢棄物處置方式，但卻導致空氣污染物的排放。為改善垃圾焚化的排放問題，我國政府提出了加嚴管制焚化爐排放污染物的政策與規範。然而，目前多數焚化爐仍無法達到加嚴管制的標準，顯示焚化爐排放管理仍需要進一步檢視與討論。此外，與要達到焚化爐減排相關的因素除了空氣污染防制設備技術的更新外，還有長期被忽視的垃圾組成因素及源頭減量政策。

本研究旨在釐清焚化爐減排的關鍵因素，以政策檢驗及機器學習模型的方式探討相關因子。首先，通過季節性分解與中斷時間序列分析，檢驗源頭減量政策及焚化爐改建政策對垃圾減量及焚化爐減排的效益。其次，採用特徵工程方法，分析減排因子對焚化爐排放的相關性與顯著性。再應用機器學習方法建立焚化爐排放預測模型，並分析輸入因子對模型的影響程度。最後，對政策檢驗、特徵工程及機器學習建模三階段的結果進行比較，以確定焚化爐減排的關鍵參數。

研究結果顯示，在政策檢驗階段，「限制塑膠吸管使用」及「雙北專用垃圾袋互收」政策對垃圾減量具有立即減量但逐漸增量的顯著效果。然而，由於空氣污染監測數據無呈現一致性，本研究未能確定該政策對焚化爐減排的效益；「新店焚化爐改建」政策顯示出空氣污染物排放值逐漸減少的顯著效果，表明該政策對焚化爐減排具有長期效益。在特徵工程階段，O2(%) 與焚化爐設計參數，特別是空氣污染物防制設備，對排放濃度影響顯著(p-value <0.001)。模型驗證結果顯示，NOx 模型(MAPE=10.4%; R2=0.714) 在穩定性及準確性上優於其他四個污染物排放預測模型， 其表現依次為HCl (MAPE=82.8%; R2=0.65) 、CO(MAPE=75.4%; R2=0.26)與SO2 (MAPE=96.6%; R2=0.39) 預測模型。SHAP 分析結果顯示，空氣污染物防制設備對四個模型的影響程度高於其他因子。對於組成參數而言，高度影響的組成參數與文獻回顧中的解釋一致。

綜合上述結果，空氣污染物防制設備仍是改善焚化爐排放的最重要且直接的因子。然而，源頭減量及垃圾組成仍然是輔助焚化爐減排的不可忽視的關鍵。

關鍵字

焚化爐 ； 空氣污染排放 ； 源頭減量 ； 中斷時間序列分析 ； 機器學習

並列摘要

As socio-economic development progresses, the amount of urban solid waste is increasing. Incineration processes are the primary waste disposal method in Taiwan, but they result in the emission of air pollutants. To address the issue of emissions from incinerators, the Taiwanese government has proposed stricter regulations and standards for incinerator emissions. However, most incinerators currently fail to meet these standards. Factors related to reducing emissions from incinerators include not only updates to air pollution control technologies but also long-neglected factors such as waste composition and source reduction policies.

This study aims to identify the key factors in reducing emissions from incinerators. First, through seasonal decomposition and interrupted time series analysis, the effects of source reduction policies and incinerator reconstruction policies on waste reduction and emission reduction were examined. Second, feature engineering methods were performed to analyze the correlation and significance of emission reduction factors on incinerator emissions. Furthermore, machine learning methods were used to establish prediction models for incinerator emissions and to analyze the impact of emission reduction factors on these models. Finally, the results from the policy examination, feature engineering, and machine learning modeling phases were compared to determine the key factors for emission reduction in incinerators.

The results indicate that during the policy examination phase, the "Restriction on Plastic Straw Usage" and "Reciprocal Use of Special Garbage Bags in Taipei and New Taipei City" policies had an immediate but gradually increasing significant effect on waste reduction. However, due to inconsistent air pollution monitoring data, this study could not confirm the policy's effectiveness in reducing emissions from incinerators. The "Xindian Incinerator Reconstruction" policy showed a gradual decrease in significant effect, indicating that the policy has long-term benefits. In the feature engineering phase, O2 (%) and incinerator design parameters, particularly air pollution control equipment, had a significant impact on emission concentrations (p-value <0.001). Model validation results show that the NOx model (MAPE=10.4%; R2=0.714) performed better in terms of stability and accuracy compared to the other four pollutant emission prediction models, followed by the HCl (MAPE=82.8%; R2=0.652), CO (MAPE=75.4%; R2=0.259), and SO2 (MAPE=96.6%; R2=0.392) prediction models.SHAP analysis results indicate that air pollution control equipment had a greater impact on the four models compared to other factors. However, for composition parameters, the highly influential parameters were consistent with explanations found in the literature and still had a significant impact.

In summary, air pollution control equipment remains the most important and direct factor in improving incinerator emissions. Nonetheless, source reduction and waste composition are also critical auxiliary factors in emission reduction.

並列關鍵字

Incinerator ； Air Pollution Emissions ； Source Reduction ； Interrupted Time Series Analysis ； Machine Learning

參考文獻

Adeleke, O., Akinlabi, S. A., Jen, T.-C., & Dunmade, I. (2021). Application of Artificial Neural Networks for Predicting the Physical Composition of Municipal Solid Waste: an Assessment of the Impact of Seasonal Variation. Waste Manag Res, 39(8), 1058-1068. https://doi.org/10.1177/0734242X21991642

Asif Afzal, Aabid, A., Khan, A., Khan, S. A., Rajak, U., Verma, T. N., & Kumar, R. (2020). Response Surface Analysis, Clustering, and Random Forest Regression of Pressure in Suddenly Expanded High-speed Aerodynamic Flows. Aerospace Science and Technology, 107, 106318.

Bergstra, J., Bardenet, R., Bengio, Y., & Kégl, B. (2011). Algorithms for Hyper-Parameter Optimization. Advances in neural information processing systems, 24.

Breiman, L. (2001). Random Forests. Machine learning, 45, 5-32.

Buekens, A., & Cen, K. (2011). Waste incineration, PVC, and dioxins. Journal of Material Cycles and Waste Management, 13, 190-197.