參加競賽名稱：2025 International Conference on Smart Devices and Sustainable Energy ( SDSE 2025 ) 論文競賽

獲得獎項：Excellence Award

題目：Enhancing COD Prediction in Aeration Tanks Through Hybrid Machine Learning and Temporal Feature Integration

同學姓名：Elysée MUGABIRE

指導教授：劉弘一

作品/競賽簡介：
劉弘一老師研究室指導之法國巴黎電子工程師高等學校（ESIEE Paris）實習生Elysée MUGABIRE，來校進行短期研究期間之研究成果，本研究建立一套基於機器學習的軟體感測器，用於污水處理曝氣池中化學需氧量（COD）的即時監測，研究採用低程式碼機器學習工具 PyCaret 的回歸模組，並引入滾動建模策略以提升預測準確性。此方法顯著提升曝氣池化學需氧量（COD）即時預測精度（RMSE 1.76, R² 0.96），並榮獲國際研討會Excellence Award肯定，展現本中心與AI學程在跨領域應用與實務研究上的卓越成效。
This study proposes machine learning-based soft sensors for real-time chemical oxygen demand (COD) monitoring in wastewater treatment aeration tanks, comparing multiple predictive modeling approaches. The analysis primarily utilizes the regression module of PyCaret, a low-code machine learning tool, and implements its rolling modeling strategy to enhance prediction accuracy. The results demonstrate that the rolling modeling strategy significantly improves the performance of regression prediction models. Additionally, incorporating time-series forecasting results of the target variable as supplementary features proves to be another critical factor in enhancing model performance. The optimal COD prediction model was developed using a Huber regressor combined with the rolling modeling strategy (window size: 250) and time-series features, achieving peak performance with an RMSE of 1.76, MAE of 1.32, and R² score of 0.96. These findings demonstrate that the hybrid modeling approach, which combines AutoML with time-series analysis, effectively captures the dynamic characteristics of COD fluctuations. This methodology enables substantial energy savings in aeration tanks – recognized as the most energy-intensive components in wastewater treatment processes.