Comparative Performance of Three Machine Learning Models in Predicting Influent Flow Rates and Nutrient Loads at Wastewater Treatment Plants

Accurately predicting influent wastewater quality is
vital for
the efficient operation and maintenance of wastewater treatment plants
(WWTPs). This study evaluated three machine learning (ML) models for
predicting influent flow rates and nutrient loads of both industrial
and domestic wastewaters in WWTPs. These predictions were based on
meteorological data and the population migration patterns. The modelsrandom
forest, extra trees, and gradient boosting regressorwere successfully
applied to three full-scale WWTPs in Shenzhen, China. All the models
demonstrated robust performance in predicting influent flow rate,
ammoniacal nitrogen (NH3–N), and total nitrogen
(TN). Feature importance analysis revealed that the average precipitation
over the past n days and population migration were
the most influential factors for predicting influent flow rate. Conversely,
human activities have a greater impact on pollutant concentrations.
Scenario analyses indicated that precipitation contributed to approximately
5%–10% of the wastewater influent, while groundwater infiltration
accounted for around 20%. Overall, this study provides a model framework
for forecasting wastewater loads using meteorological and population
migration data, setting the groundwork for smart management in WWTPs.