基于树集成模型的污水处理工艺亚硝酸盐氮软测量研究

马亚朋; 李壮壮; 徐敬生; 吕路

doi:10.13205/j.hjgc.202504012

基于树集成模型的污水处理工艺亚硝酸盐氮软测量研究

doi: 10.13205/j.hjgc.202504012

马亚朋¹,
李壮壮^1,2,
徐敬生²,
吕路^1,2, ,

1. 南京大学环境学院污染控制与资源化研究国家重点实验室,南京 210046;
2. 南京大学常高新国际环保产业技术研究院,江苏常州 213125

基金项目:

国家自然科学基金重点项目“面向深度水处理的限域纳米铁氧化物复合材料构建、特性与应用”（22236003）

详细信息

作者简介:
马亚朋（1999—），男，硕士研究生，主要研究方向为污水处理工艺智慧化管控。mg21250056@smail.nju.edu.cn

通讯作者:
吕路（1975—），男，教授，主要研究方向为大数据与工业污染减排过程控制。esellu@nju.edu.cn

计量
- 文章访问数: 76
- HTML全文浏览量: 31
- PDF下载量: 2
- 被引次数: 0
出版历程
- 收稿日期: 2023-11-08
- 录用日期: 2024-03-28
- 修回日期: 2024-01-18
- 刊出日期: 2025-04-01

Research on soft sensing of nitrite nitrogen in wastewater treatment process based on tree integration models

1. State Key Laboratory of Pollution Control and Resource Reuse, School of Environment, Nanjing University, Nanjing 210046, China;
2. Joint Institute of International Environmental Technology Innovation of Changzhou Hi-Tech District and Nanjing University, Changzhou 213125, China

摘要

摘要: 机器学习在污水处理工艺水质监测的应用成为当前研究热点。为解决关键水质指标的获取存在时间滞后性和成本高的问题，以短程硝化反硝化工艺亚硝酸盐氮（NO₂^--N）浓度预测为目标，提出了一种基于树集成模型的水质软测量方法。采用反硝化池易监测的参数，应用决策树的四类集成模型预测出水NO₂^--N浓度，通过分析输入特征对预测结果的重要性程度来解释最优模型，并用特征选择验证模型解释结果。结果表明，4种决策树集成模型中，自适应提升（AdaBoost）对于出水NO₂^--N浓度的预测准确率和稳定性最高，决定系数（R²）、均方误差（MSE）和平均绝对百分比误差（MAPE）分别为0.983、0.015和0.126；模型解释表明，pH、ORP及进水COD是影响预测效果的高重要性参数，与NO₂^--N呈现出较强相关性。该研究对于实现低成本、实时地预测水质指标，扩充有效数据量提高预测精度具有重要参考价值。
- 机器学习 /
- 树集成模型 /
- 水质软测量 /
- 短程硝化反硝化 /
- 模型解释
Abstract: The application of machine learning in the water quality monitoring of wastewater treatment processes has become a significant focus of contemporary research. Addressing the challenges about time lag and high detection cost in acquiring key water quality indicators of traditional monitoring methods in wastewater treatment processes， this study focused on the prediction of nitrite nitrogen （NO₂^--N） concentration in the shortcut nitrification-denitrification process， proposed a novel soft sensing method for water quality indicators based on the decision tree integration model. This study selected some easily measured and important operating parameters as the input features of the water quality soft sensing method， constructed four categories of tree integration models based on the decision tree to predict the effluent NO₂^--N concentration accurately in the denitrification process. The prediction accuracy and stability of each model were compared to select the best-performing one. The importance level of these models’ input features to the NO₂^--N prediction results was analyzed to interpret the optimal prediction model. Moreover， the study utilized feature selection analysis to further validate the interpretability of these tree integration models’ results. The research results indicated that among these four decision tree integration models， the Adaptive Boosting （AdaBoost） model exhibited the highest predictive accuracy and stability for the effluent NO₂^--N concentration in the denitrification process. In the prediction performance of the AdaBoost model， the determination coefficient （R²）， mean square error （MSE） and mean absolute percentage error （MAPE） was 0.983， 0.015 and 0.126， better than other models. The results of model interpretation revealed that pH， oxidation reduction potential （ORP） and influent chemical oxygen demand （COD） concentration were the most critical parameters for these decision tree integration models. These parameters affected models’ prediction results significantly and exhibited strong correlations with the effluent NO₂^--N concentration in the denitrification process. Further feature selection analysis confirmed that these parameters play crucial roles in improving the prediction accuracy of these models. In this study， the soft sensing method based on decision tree integration models provided a valuable reference for achieving low-cost and real-time prediction of water quality indicators， expanding the availability of effective data to enhancing prediction accuracy for all kinds of prediction models in wastewater treatment processes.
- machine learning /
- tree integration model /
- soft sensing of water quality /
- shortcut nitrification-denitrification /
- model interpretation

HTML全文

1 污水处理工艺流程

1. Sewage treatment process flow

下载: 全尺寸图片幻灯片

2 小试工艺流程和数据监测点位

2. Flow path and data monitoring points of the small-scale test process

下载: 全尺寸图片幻灯片

3 污水处理工艺各参数随时间的变化

3. The parameters in the wastewater treatment process varied with time

下载: 全尺寸图片幻灯片

4 基于树集成模型的出水NO2--N浓度软测量算法流程

4. Algorithm flow of effluent NO2--N concentration soft sensing based on tree integration models

下载: 全尺寸图片幻灯片

5 输入输出参数间的Pearson相关系数结果

5. Pearson correlation coefficient between input and output parameters

下载: 全尺寸图片幻灯片

6 7个输入特征与出水NO2--N的线性拟合结果

6. Linear fitting results of seven input features and effluent NO2--N concentrations

下载: 全尺寸图片幻灯片

7 基础算法构建出水NO2--N预测模型的结果

7. The results of the effluent NO2--N prediction models established by basic algorithms

下载: 全尺寸图片幻灯片

8 树集成算法构建出水NO2--N预测模型的结果

8. The results of the effluent NO2--N prediction models established by tree integration algorithms

下载: 全尺寸图片幻灯片

9 树集成模型在测试集上预测值和实际值的对比

9. Comparisons of predicted and true values of tree integration models in the test set

下载: 全尺寸图片幻灯片

10 树集成模型在测试集上的预测准确率分布

10. Prediction accuracy distribution of tree integration models in the test set

下载: 全尺寸图片幻灯片

11 AdaBoost和XGBoost的置换特征重要性参数

11. Permutation importance factors of AdaBoost and XGBoost

下载: 全尺寸图片幻灯片

12 AdaBoost和XGBoost的平均绝对SHAP值

12. Mean absolute SHAP values of AdaBoost and XGBoost

下载: 全尺寸图片幻灯片

13 从低到高重要性依次剔除输入参数对模型预测效果的影响

13. Impact of removing input variables by ascending order of importance on model prediction result

下载: 全尺寸图片幻灯片

14 从高到低重要性依次剔除输入参数对模型预测效果的影响

14. Impact of removing input variables by decreasing order of importance on model prediction result

下载: 全尺寸图片幻灯片

1 不同模型的最佳超参数值

1. The optimal hyperparameters values of different models

模型	超参数	最适模型超参数
RF	n_estimators	200
	max_depth	10
	max_features	3
	max_samples	0.9
	min_samples_split	2
Adaboost	n_estimators	100
	learning_rate	2.1
	subsample	1.0
	loss	exponential
	min_samples_split	2
GBRT	n_estimators	200
	learning_rate	0.2
	subsample	1.0
	max_features	3
XGBoost	n_estimators	500
	learning_rate	0.05
	subsample	0.8
	min_child_weight	3
	max_depth	7

下载: 导出CSV

2 基础及树集成模型的预测性能指标

2. Predictive performance indicators of basic and tree integration models

模型	训练集R²	测试集R²	训练集MSE	测试集MSE	训练集MAPE	测试集MAPE
线性回归	0.715	0.675	0.286	0.324	0.523	0.545
决策树	0.949	0.936	0.030	0.053	0.209	0.260
随机森林	0.981	0.975	0.021	0.024	0.115	0.149
自适应提升	0.988	0.983	0.012	0.015	0.097	0.126
梯度提升回归树	0.977	0.974	0.023	0.025	0.159	0.178
极端梯度提升	0.986	0.980	0.012	0.019	0.103	0.155

下载: 导出CSV

参考文献(33)

[1]	LEWIS W，WURTSBAUGH W，PAERL H，et al. Rationale for control of anthropogenic nitrogen and phosphorus to reduce eutrophication of inland waters［J］. Environmental Science& Technology，2011，45（24）:10300-10305.
[2]	MORRIS L，COLOMBO V，HASSELL K，et al. Municipal wastewater effluent licensing:a global perspective and recommendations for best practice［J］. Science of the Total Environment，2017，580:1327-1339.
[3]	XUE T L，ZHAO D H，HAN F. Svr water quality prediction model based on GA optimization［J］. Environment Engineering，2020，38（3）:123-127. 薛同来，赵冬晖，韩菲. 基于GA优化的SVR水质预测模型研究［J］. 环境工程，2020，38（3）:123-127.
[4]	ZHONG S，ZHANG K，BAGHERI M，et al. Machine learning:new ideas and tools in environmental science and engineering［J］. Environmental Science& Technology，2021，55（19）:12741-12754.
[5]	LI K，DUAN H，LIU L，et al. An integrated first principal and deep learning approach for modeling nitrous oxide emissions from wastewater treatment plants［J］. Environmental Science& Technology，2022，56（4）:2816-2826.
[6]	WANG G，JIA Q，ZHOU M，et al. Artificial neural networks for water quality soft-sensing in wastewater treatment:a review［J］. Artificial Intelligence Review. 2022； 55（1）:565-587.
[7]	LI Y Y，LIU H L. Prediction of total phosphorus in rivers based on attention mechanism of temporal convolutional networks［J］. Environment Engineering，2023，41（5）:163-171. 黎园园，刘海隆. 基于注意力机制的时间卷积网络河流总磷预测［J］. 环境工程，2023，41（5）:163-171.
[8]	DU R，PENG YZ. Technical revolution of biological nitrogen removal from municipal wastewater:recent advances in anammox research and application［J］. Scientia Sinica Technologica，2022，52（3）:389-402. 杜睿，彭永臻. 城市污水生物脱氮技术变革:厌氧氨氧化的研究与实践新进展［J］. 中国科学:技术科学，2022，52（3）:389-402.
[9]	ZHANG L，JIANG L，ZHANG J，et al. Enhancing nitrogen removal through directly integrating anammox into mainstream wastewater treatment:Advantageous，issues and future study［J］. Bioresour Technol，2022，362:127827.
[10]	LI J，WU L N，YAN Z B，et al. Research on control strategy of nitrogen removal functional bacteria in mainstream partial nitrification-anammox process［J］. Environment Engineering，2021，39（8）:45-54，61. 李进，吴莉娜，闫志斌，等. 主流部分亚硝化-厌氧氨氧化工艺脱氮功能菌群的控制策略研究［J］. 环境工程，2021，39（8），45-54，61.
[11]	WU P，CHEN J，GARLAPATI V，et al. Novel insights into Anammox-based processes:A critical review［J］. Chemical Engineering Journal，2022，13（2）:444.
[12]	AGRAWAL S，SEUNTJENS D，COCKER P，et al. Success of mainstream partial nitritation/anammox demands integration of engineering，microbiome and modeling insights［J］. Current Opinion in Biotechnology，2018，50:214-221.
[13]	GUO L J，LI B L. Research on soft measurement of effluent total nitrogen based on GNFA-SVR［J］. Industrial Water Treatment，2022，42（10）:111-117. 郭利进，李博仑. 基于GNFA-SVR污水出水总氮软测量研究［J］. 工业水处理，2022，42（10）:111-117.
[14]	ZHANG M，YANG Y，GUO W，et al. Electrochemical sensor for sensitive nitrite and sulfite detection in milk based on acid-treated Fe₃O₄@SiO₂ nanoparticles［J］. Food Chemistry，2024，35（6）:430.
[15]	LI X M，LUO X K，LI W，et al. Design of a nitrite sensor based on colorimetry［J］. Journal of Beijing University of Chemical Technology（Natural Science Edition），2020，47（6）:79-84. 李新民，罗学科，李文，等. 基于颜色检测的亚硝酸盐传感器设计［J］. 北京化工大学学报（自然科学版），2020，47（6）:79-84.
[16]	WANG C Z，WANG X C，NIU Q，et al. Assessment on measurement uncertainty of nitrite nitrogen in water by n-（1-naphthyl）-ethylenediamine spectrometry［J］. Environment Engineering，2013，31（5）:147-150. 王存政，王兴春，牛倩，等. N-（1-萘基）-乙二胺光度法测定水中亚硝酸盐氮的测量不确定度评定［J］. 环境工程，2013，31（5）:147-150.
[17]	MENG X，ZHANG Y，QIAO J，et al. An adaptive task-oriented RBF network for key water quality parameters prediction in wastewater treatment process［J］. Neural Computing& Applications，2021，33（17）:11401-11414.
[18]	FOSCHI J，TUROLLA A，ANTONELLI M，et al. Soft sensor predictor of E. coli concentration based on conventional monitoring parameters for wastewater disinfection control［J］. Water Research，2021，14（9）:191.
[19]	ZHU J，JIANG Z，FENG L，et al. Improved neural network with least square support vector machine for wastewater treatment process［J］. Chemosphere，2022，15（1）:308.
[20]	GORGAN F，RAJAEE T，ZOUNEMAT M，et al. Investigating machine learning models in predicting lake water quality parameters as a 3-year moving average［J］. Environmental Science and Pollution Research，2023，30（23）:63839-63863.
[21]	QUINLAN J. Improved use of continuous attributes in C4.5［J］. Journal of Artificial Intelligence Research，1996，4:77-90.
[22]	RUDIN C. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence［J］. 2019，1（5）:206-215.
[23]	WANG D，THUNELL S，LINDBERG U，et al. A machine learning framework to improve effluent quality control in wastewater treatment plants［J］. Science of the Total Environment，2021，3（9）:784.
[24]	HE X，JI J，LIU K，et al. Soft sensing of silicon content via bagging local semi-supervised models［J］. Sensors，2019，19（17）:36-45.
[25]	WANG R，YU Y，CHEN Y，et al. Model construction and application for effluent prediction in wastewater treatment plant:Data processing method optimization and process parameters integration［J］. Journal of Environmental Management，2022，9（14）:302-308.
[26]	LI X，YI X，LIU Z，et al. Application of novel hybrid deep leaning model for cleaner production in a paper industrial wastewater treatment system［J］. Journal of Cleaner Production，2021，14（6）:294-301.
[27]	LIU X，SHANG J，SHI H，et al. Research on Real-Time Data Processing Method of GNSS-RTK Deformation Monitoring［J］. Chinese Journal of Sensors and Actuators，2020，33（8）:1190-1196.
[28]	DOU J，YUNUS A，BUI D，et al. Improved landslide assessment using support vector machine with bagging，boosting，and stacking ensemble machine learning framework in a mountainous watershed，Japan［J］. Landslides，2020，17（3）:641-658.
[29]	BREIMAN L. Pasting small votes for classification in large databases and on-line［J］. Machine Learning，1999，36（1/2）:85-103.
[30]	CHENG W，YUAN D，XIONG P，et al. Construction and evaluation of city water quality index prediction model based on multiple machine learning algorithms［J］. Acta Scientiae Circumstantiae，2023，43（11）:144-152.
[31]	BIAN L，QIN X，ZHANG C，et al. Application，interpretability and prediction of machine learning method combined with LSTM and LightGBM-a case study for runoff simulation in an arid area［J］. Journal of Hydrology，2023，4（13）:625.
[32]	PARK J，LEE W，KIM K，et al. Interpretation of ensemble learning to predict water quality using explainable artificial intelligence［J］. Science of the Total Environment，2022，19（7）:832-839.
[33]	LI Y，WANG Y，DONG F，et al. Controlling carbon dioxide-to-hydrogen ratio to improve hydrogen utilization and denitrification rates of hydrogenotrophic autotrophic denitrification through homoacetogenesis-heterotrophic denitrification pathway［J］. Bioresource Technology，2024，16（7）:393-400.