Cloud-Model-Based Feature Engineering to Analyze the Energy–Water Nexus of a Full-Scale Wastewater Treatment Plant

Prediction and Evaluation of Indirect Carbon Emission from Electrical Consumption in Multiple Full-Scale Wastewater Treatment Plants via Automated Machine Learning-Based Analysis

“…In this study, 13 input variables, including the category of STP, category of TTP, IV, influent COD, influent NH 4 + -N, influent TN, influent TP, influent SS, effluent COD, effluent NH 4 + -N, effluent TN, effluent TP, and effluent SS, were utilized to predict indirect carbon emissions from electrical consumption of WWTPs. These input variables were chosen due to their close relationships with electrical consumption and easy availability. , The data of numerical input variables were first normalized by Z-score normalization, whereas the categorical variables ( i.e. , STP and TTP) were encoded by three types of category encoders, including ordinal encoder (OE), leave-one-out encoder (LOOE), and quantile encoder (QE) (Tables S2 and S3).…”

“…Results showed that the mean specific electrical consumptions of W9, which consisted of the SBR and cloth-media filter (low-efficient processes identified by the dependence plots), were extremely higher than the other three WWTPs. Although W5 had a lower mean specific electrical consumption for COD (1.43 kWh/kg COD) removal than W3 (2.68 kWh/kg COD) and W4 (2.20 kWh/kg COD), the mean specific electrical consumption of W5 was still larger than the data (0.76−1.01 kWh/kg COD) reported by Yang et al 2 Therefore, further optimizations are required to improve the energy efficiency of W5 and W9.…”

“…These input variables were chosen due to their close relationships with electrical consumption and easy availability. 2,24 The data of numerical input variables were first normalized by Z-score normalization, whereas the categorical variables (i.e., STP and TTP) were encoded by three types of category encoders, including ordinal encoder (OE), leave-one-out encoder (LOOE), and quantile encoder (QE) (Tables S2 and S3). OE uses a single column of integers to represent the classes, whereas LOOE excludes the current row's target when calculating the mean target for a level to reduce the effect of outliers.…”

“…Wastewater treatment plants (WWTPs) have been considered as energy- and material-intensive fields with high carbon emissions, contributing approximately 3% of total human activity emissions. − Therefore, the reduction of carbon emissions in WWTPs contributed greatly to global carbon neutrality . Nowadays, intensive attention has been focused on direct carbon emissions ( i.e.…”

“…As a complex system, the electrical consumption of WWTPs is influenced by diverse factors, such as treatment processes ( e.g. , different primary, secondary and tertiary treatment processes), influent quality, and detailed operation strategies, ,− which is time-consuming and complex to disclose their inter-relations manually. With the rapid development of machine learning (ML), some studies have established ML models for modeling the electrical consumption of WWTPs.…”