好氧-缺氧-好氧（OAO）生物處理程序操作參數融入智慧化系統之可行性研究;A feasibility study on integrating the operation parameters of aerobic-anoxic-aerobic (OAO) biological wastewater treatment process into an intelligent system

NCU Institutional Repository > 工學院 > 環境工程研究所 > 博碩士論文 > Item 987654321/98685

請使用永久網址來引用或連結此文件: https://ir.lib.ncu.edu.tw/handle/987654321/98685

題名:	好氧-缺氧-好氧（OAO）生物處理程序操作參數融入智慧化系統之可行性研究;A feasibility study on integrating the operation parameters of aerobic-anoxic-aerobic (OAO) biological wastewater treatment process into an intelligent system
作者:	高子翔;Kao, Zi-Xiang
貢獻者:	環境工程研究所
關鍵詞:	OAO除氮程序;水力停留時間;污泥停留時間;時序模型;RF;預測
日期:	2025-07-24
上傳時間:	2025-10-17 13:05:04 (UTC+8)
出版者:	國立中央大學
摘要:	隨著工業化與都市化快速發展，大量含有機物與氮的污染物排入水體，導致淡水生態優養化問題日益嚴重，威脅水生生物與人類健康。國內放流水標準日益嚴格，使污水處理廠面臨更高的處理要求與符合法規並降低對環境衝擊，深入探討處理機制與優化策略已成當務之急。傳統經驗式操作已難以應對複雜的處理條件與標準，導入人工智慧與機器學習等技術，助於系統升級。透過即時數據監測與模型建構，可預測水質變化趨勢，優化操作參數，提升系統穩定性與效能，並強化放流水管理與決策準確性。本研究於桃園北區水資源中心設置模型廠，探討好氧-缺氧-好氧（OAO）生物除氮程序之不同操作條件如：水力停留時間（HRT）、污泥停留時間（SRT）和有機負荷（F/M）等關鍵參數下之處理效能，尋求污染物去除條件之極限。試驗過程將分別調整HRT由15縮短至10小時，ASRT由15降至10天，三階段皆展現良好SS去除率（88.9–90.1%）， TCOD與BOD去除率穩定維持在83–89%，即便在高F/M條件下（0.21±0.11 kg BOD/kg MLSS∙d）仍具穩定有機物處理效能。氨氮去除率則由87.0%降至75.8%，顯示 HRT 與 ASRT 縮短對硝化反應造成影響。提升溶氧至1.81±1.53 mg/L後，氨氮去除率回升至90.5%，但亦造成ORP變動加劇，影響脫硝穩定性。整體結果顯示，本系統於HRT 10–13小時、SRT 10天條件下仍具處理潛力。另外，本研究結合智慧化技術與人工智慧（Artificial Intelligence , AI）模型，應用時序模型（Transient reaction model , TRM）與隨機森林（Random Forest , RF）模型，結合水質感測器的即時數據與操作條件等，預測五項水質參數（TCOD、SS、NH₄⁺-N、NO₃⁻-N、pH）。缺失值處理比較，顯示線性內插較能保留時間序列且具工程意義之預測參數排序，適合建模使用。在模型預測效能方面，水質模型預測過程可以觀察到單一輸入太過單薄，增加輸入參數有助於預測結果大幅改善；而以不同參數組合結果中表明，本研究之三種不同組合分別代表基礎資訊（監測參數）、環境條件（狀態參數）與調控策略（控制參數），以不同數據組合可以表明不同資訊對之水質結果之影響程度。其中TCOD與SS預測分別達到R2 = 0.9943、 RMSE = 0.8790 mg/L 與 R2 = 0.9941、RMSE = 0.7347 mg/L，顯示 RF 在面對相對穩定之物理或有機性指標時具備高準確性與低誤差的優勢。時序模型在結合監測與控制參數後，於氮類水質參數如NH4⁺-N與NO3⁻-N表現更佳，預測R2分別為0.9943 與 0.9875，RMSE 分別為0.2602與0.3883 mg/L，凸顯時序模型對於動態變異性高之參數具強化預測效能的能力。綜合分析，RF 模型適用於靜態參數與即時監測應用，而時序模型則適合掌握與操作策略變動相關之複雜動態行為。建議未來依據應用需求選擇適當模型與參數組合，並持續累積高頻率之實測與控制資料，以進一步發展智慧預警與動態決策支援系統，支持自適應性管理，提升污水處理效能與管理效率，為因應嚴格排放標準與環境挑戰提供技術支援。本研究將OAO程序操作參數最佳化，並整合至智慧系統中，提升廢水處理中的脫氮效率與除碳穩定性，強化管理策略，兼顧成本與環保效益。;With the rapid development of industrialization and urbanization, large amounts of organic and nitrogen-containing pollutants are being discharged into water bodies, leading to increasingly severe eutrophication of freshwater ecosystems and posing threats to aquatic life and human health. As effluent discharge standards in Taiwan become more stringent, wastewater treatment plants are facing higher treatment demands. To comply with regulations and reduce environmental impacts, it is imperative to explore treatment mechanisms and optimization strategies in greater depth. Traditional experience-based operations can no longer meet the complexity of modern treatment conditions and standards; therefore, the integration of artificial intelligence（AI） and machine learning technologies has become key to system upgrades. Through real-time data monitoring and model development, it is possible to predict water quality trends, optimize operational parameters, enhance system stability and efficiency, and improve effluent management and decision-making accuracy. This study established a pilot-scale plant at the Taoyuan Northern Water Resources Recycling Center to investigate the treatment performance of the Oxic-Anoxic-Oxic （OAO）biological nitrogen removal process under various operational conditions, including key parameters such as hydraulic retention time（HRT）, sludge retention time（SRT）, and organic loading rate（F/M）. The aim was to identify the limits of pollutant removal efficiency. During the experiments, the HRT was reduced from 15 to 10 hours and the aerobic sludge retention time（ASRT）from 15 to 10 days. Across all three test stages, the system consistently achieved high suspended solids（SS） removal efficiency（88.9–90.1%）and stable removal of TCOD and BOD（83–89%）, even under high F/M conditions（0.21±0.11 kg BOD/kg MLSS∙d）, indicating effective organic matter treatment. However, the ammonia nitrogen（NH4⁺-N）removal efficiency declined from 87.0% to 75.8%, suggesting that shortened HRT and ASRT adversely affected the nitrification process. After increasing the dissolved oxygen（DO）concentration to 1.81±1.53 mg/L, NH4⁺-N removal improved to 90.5%. Nonetheless, the elevated DO caused greater fluctuations in oxidation-reduction potential（ORP）, impacting the stability of the denitrification process. Overall, the results demonstrate that the system retains treatment potential under HRT conditions of 10–13 hours and an SRT of 10 days. This study integrates intelligent technologies and artificial intelligence（AI）models by applying a time-series model（TCR）and a Random Forest（RF）model to predict five water quality parameters—TCOD, SS, NH4⁺-N, NO3⁻-N, and pH—based on real-time data from water quality sensors and operational conditions. A comparison of missing value treatments showed that linear interpolation better preserves time-series continuity and provides engineering-relevant parameter ordering, making it suitable for model construction. In terms of model performance, the results indicate that using only a single input variable yields limited predictive power, while incorporating additional input parameters significantly improves prediction accuracy. The three input combinations adopted in this study—monitoring parameters（basic information）,status parameters （environmental conditions）, and control parameters（operational strategies） —demonstrate how different types of data influence water quality prediction outcomes. Among the results, TCOD and SS predictions achieved R2 values of 0.9943 and 0.9941, with RMSE values of 0.8790 mg/L and 0.7347 mg/L, respectively. These results highlight RF’s advantage in predicting relatively stable physical or organic indicators with high accuracy and low error. On the other hand, the time-series model, when combined with monitoring and control parameters, performed better for nitrogen-related parameters. For NH4⁺-N and NO3⁻-N, the time-series model achieved R2 values of 0.9943 and 0.9875, and RMSE values of 0.2602 and 0.3883 mg/L, respectively, demonstrating its enhanced predictive capability for dynamic variables. In summary, the RF model is suitable for predicting static parameters and real-time monitoring applications, while the time-series model excels at capturing complex dynamic behaviors related to changes in operational strategies. It is recommended that future applications select appropriate models and input combinations based on specific needs, and continuously accumulate high-frequency monitoring and control data to further develop intelligent early-warning and dynamic decision support systems. Such systems can support adaptive management, enhance treatment performance, and improve operational efficiency, thereby providing technical support to meet increasingly stringent discharge standards and environmental challenges. This study optimized the operational parameters of the OAO process and integrated them into an intelligent system to enhance nitrogen removal efficiency and carbon removal stability in wastewater treatment, strengthen management strategies, and achieve both cost-effectiveness and environmental benefits.
顯示於類別:	[環境工程研究所 ] 博碩士論文

文件中的檔案:

檔案	描述	大小	格式	瀏覽次數
index.html		0Kb	HTML	31	檢視/開啟

在NCUIR中所有的資料項目都受到原著作權保護.

社群 sharing

資料載入中.....