摘要
生化需氧量(BOD)是污水处理厂水质监测的重要指标。污水处理厂进水BOD指标的传统检测方法存在测试时间长、实际操作复杂等局限性,无法为污水处理厂根据进水水质波动进行工艺参数调整提供及时和准确的指导。为此,研究了使用支持向量机回归(Support Vector Regression, SVR)及极端梯度提升(Extreme Gradient Boosting, XGBoost)两种机器学习模型算法对进水BOD进行软测量的可行性,同时辅以遗传算法(Genetic Algorithm, GA)进行参数优化。结果表明,采用GA进行参数优化能够提升SVR模型与XG-Boost模型的预测精度,均方根误差(Root Mean Squared Error, RMSE)分别降低10.70%与33.33%。相比较GA-XGBoost模型,使用GA-SVR方法进行预测的结果更准确,拟合度(R2)可达0.918。研究结果可为污水处理厂进水BOD指标软测量方法的工程应用提供数据支撑。
Biochemical oxygen demand(BOD)is an important indicator of water quality monitoring in wastewater treatment plants.Owing to the limitations of long test time and high practical operation requirements,the traditional detection method of BOD index cannot provide timely and accurate guidance to adjust process parameters according to the fluctuation of influent water quality in wastewater treatment plant.Hence,this study investigated the feasibility of soft measurement of BOD by using two machine learning model algorithms,Support Vector Regression(SVR)and Extreme Gradient Boosting(XGBoost).Meanwhile,Genetic Algorithm(GA)is used to optimize the parameters.The results show that GA can improve the prediction accuracy of SVR and XGBoost,and the root mean square error(RMSE)can be reduced by 10.70%and 33.33%respectively.Compared with GA-XGBoost,the predicted result by GA-SVR is more accurate and its R-squared can reach 0.918.The research results of this paper can provide data support for engineering applications of BOD soft measurement in wastewater treatment plants.
作者
苗露
姚怡帆
王黎佳
王丽艳
黄黎明
刘长青
MIAO Lu;YAO Yifan;WANG Lijia;WANG Liyan;HUANG Liming;LIU Changqing(Qingdao Tuandao Wastewater Treatment Plant,Qingdao 266002,China;School of Environmental&Municipal Engineering,Qingdao University of Technology,Qingdao 266525,China;Qingdao Zhangcun River Water Co.Ltd.,Qingdao 266100,China)
出处
《青岛理工大学学报》
CAS
2023年第2期133-139,共7页
Journal of Qingdao University of Technology
基金
国家重点研发计划(2020YFD1100303)。
关键词
生化需氧量(BOD)
软测量
遗传算法
机器学习
biochemical oxygen demand(BOD)
soft sensing
genetic algorithm
machine learning
