ORIGINAL RESEARCH
Influent Quality and Quantity Prediction in Wastewater Treatment Plant: Model Construction and Evaluation
Rui Wang 1, 2, 3  
,   Zhicheng Pan 4  
,   Yangwu Chen 2, 3  
,   Zhouliang Tan 2, 3  
,   Jianqiang Zhang 1  
 
More details
Hide details
1
Faculty of Geosciences and Environmental Engineering, Southwest Jiaotong University, 611756 Chengdu, China
2
Key Laboratory of Environmental and Applied Microbiology, Chengdu Institute of Biology, Chinese Academy of Sciences, 610041 Chengdu, China
3
Environmental Microbiology Key Laboratory of Sichuan Province, Chengdu Institute of Biology, Chinese Academy of Sciences, 610041 Chengdu, China
4
School of Environment, Tsinghua University,100091 Beijing, China
CORRESPONDING AUTHOR
Zhouliang Tan   

Key Laboratory of Environmental and Applied Microbiology, Chengdu Institute of biology, Chinese Academy of Sciences, No.9, 4th Block, People's South Street, Chengdu, C, 610041, Chengdu, China
Submission date: 2020-12-03
Final revision date: 2021-01-22
Acceptance date: 2021-01-25
Online publication date: 2021-06-29
Publication date: 2021-07-29
 
Pol. J. Environ. Stud. 2021;30(5):4267–4276
 
KEYWORDS
TOPICS
ABSTRACT
Influent quality and quantity were important factors that caused the abnormal operation of WWTP. In this study, the prediction models of influent quality and quantity were established based on four machine learning methods of Linear Regression, Ridge Regression, ElasticNet Regression and Lasso Regression. The meteorological conditions (precipitation and air temperature) and influent indicators (influent quantity, COD, and NH3-N) were used as training data. The influent quantity prediction of the models were evaluated using the historical data obtained from a WWTP located in western China, and the results showed that the normal rates of influent quantity were ranged from 98.9%-100%. The highest accuracy was obtained with Ridge method which was 86.19% .
For influent quality (COD) prediction, Ridge method is relatively ideal, with 82% accuracy. For influent quality (NH3-N) prediction, because of higher data normality rates, Lasso and ElasticNet method were more ideal, both with 74% accuracy. Further, in view of the reason of low prediction accuracy, this paper puts forward the idea of model improvement from the three directions of data fluctuation, correlation and amount. It is expected that this study will provide reference for similar research and provide a reference and thought for similar research.
eISSN:2083-5906
ISSN:1230-1485