應用機械學習預測污水廠放流水之氨氮

Prediction of Ammonia in Effluent of Wastewater Treatment Plant by Mechanical Learning

洪一丁(Yi-Ding Hong)

指導教授 ： 駱尚廉

國立臺灣大學/工學院/環境工程學研究所/碩士(2021年)

https://doi.org/10.6342/NTU202101330

摘要

廢污水中的氨氮是水體常見污染物，也是現今污水廠需要去除的主要物質之一。透過預測放流水氨氮，可用於輔助工作人員之最佳化操作，降低污水廠運作成本。因此，本研究將使用機械學習模型對放流水水中的氨氮進行預測，計算預測結果與實際量測值差異，選擇最適合的模型。
本研究使用迪化污水處理廠 2020 年一月至十月之每小時水質數據，使用氫離子濃度指數、水溫、導電度、化學需氧量、氨氮及懸浮固體為原始資料，經由特徵值篩選後，透過 XGBoost、梯度提升機模型、LightGBM、隨機森林模型、極度隨機樹五種機械學習模型，分別對十一月第一週放流水之氨氮進行預測。得到訓練結果後進行參數調整以優化模型，最後將訓練集和驗證集數據整合，得到最終模型。結果顯示，五種模型之準確率分別為 84.8%、40.8%、70.8%、85% 和 40%，其中 XGBoost 和隨機森林模型具有較好的預測準確率，梯度提升機和極度隨機樹模型的評鑒指標與前二者差距不大，但預測結果並不理想，推測與輸入數據的雜訊有關。此外，峰值的預測上，XGBoost 模型有更好的效果。研究表明，使用 XGBoost 等機器學習模型可在一定程度上預測放流水之氨氮濃度。

關鍵字

氨氮 ； 水質預測 ； XGBoost ； LightGBM ； 隨機森林模型

並列摘要

Ammonia in waste water is a common pollutant in water bodies, and it is also one of the main substances that need to be removed from wastewater treatment plant today. By predicting the ammonia in the effluent, it can be used to assist the staff in the optimal operation and reduce the operating cost of the wastewater treatment plant. Therefore, this study will use the mechanical learning model to predict the ammonia in the effluent, calculate the difference between the predicted result and the actual measured value, and select the most suitable model.
This study uses the hourly water quality data of Dihua Waste Water Treatment Plant from January to October 2020, using pH, water temperature, conductivity, COD, ammonia and suspended solids as the original data. After filtering by eigenvalues, through five mechanical learning models: XGBoost, Gradient Boosting Machine Model(GBM), LightGBM, Random Forest Model(RF), and Extreme Random Tree(ET), respectively, the ammonia in the effluent released in the first week of November was predicted. After the training results were obtained, the parameters were adjusted to optimize the model, and finally the training and validation data were integrated to obtain the final model. The results show that the accuracy rates of the five models are 84.8%, 40.8%, 70.8%, 85%, and 40%, respectively. Among them, XGBoost model and RF model have better prediction accuracy. The evaluation indicators of the GBM model and the ET model are not much different from the former two, but the prediction results are not ideal, and it is speculated that it is related to the noise of the input data. In addition, the XGBoost model has better results in peak prediction. Studies have shown that the use of machine learning models such as XGBoost can predict the ammonia concentration of the effluent to a certain extent.

並列關鍵字

Ammonia ； Water quality prediction ； XGBoost ； LightGBM ； Random Forest model

參考文獻

Bagherzadeh, F., Mehrani, M.-J., Basirifard, M., and Roostaei, J. (2021). Comparative study on total nitrogen prediction in wastewater treatment plant and effect of various feature selection methods on machine learning algorithms performance. Journal of Water Process Engineering, 41, 102033. doi: https://doi.org/10.1016/j.jwpe.2021.102033

Breiman, L., Friedman, J., Stone, C. J., and Olshen, R. A. (1984). Classification and regression trees: CRC press.

Chen, T., and Guestrin, C. (2016). Xgboost: A scalable tree boosting system. Paper presented at the Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining.

Del Moro, G., Barca, E., De Sanctis, M., Mascolo, G., and Di Iaconi, C. (2016). Gross parameters prediction of a granular-attached biomass reactor by means of multi-objective genetic-designed artificial neural networks: touristic pressure management case. Environmental Science and Pollution Research, 23(6), 5549-5565.

Fan, J., Ma, X., Wu, L., Zhang, F., Yu, X., and Zeng, W. (2019). Light Gradient Boosting Machine: An efficient soft computing model for estimating daily reference evapotranspiration with local and external meteorological data. Agricultural Water Management, 225, 105758. doi: https://doi.org/10.1016/j.agwat.2019.105758