Abstract:Water pollution generated from intensive anthropogenic activities has emerged as a critical issue concerning ecosystem balance and livelihoods worldwide. Although optimizing wastewater treatment efficiency is widely regarded as the foremost step to minimize pollutants released into the environment, this widespread application has encountered two major problems: firstly, the significant variation of influent wastewater constituents; secondly, complex treatment processes within wastewater treatment plants (WWTPs). Based on the data collected hourly using real-time sensors in three different full-scale WWTPs (24 h × 365 days × 3 WWTPs × 10 wastewater parameters), this work introduced the potential application of Machine Learning (ML) to predict wastewater quality. In this work, six different ML algorithms were examined and compared, varying from shallow to deep learning architectures including Seasonal Autoregressive Integrated Moving Average (SARIMAX), Random Forest (RF), Support Vector Machine (SVM), Gradient Tree Boosting (GTB), Adaptive Neuro-Fuzzy Inference System (ANFIS) and Long Short-Term Memory (LSTM). These models were developed to detect total phosphorus in the outlet (Outlet-TP), which served as an output variable due to the rising concerns about the eutrophication problem. Irrespective of WWTPs, SARIMAX consistently demonstrated the best performance for regression estimation as evidenced by the lowest values of Mean Square Error (MSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and the highest coefficient of determination (R2). In terms of computation efficiency, SARIMAX exhibited acceptable time computation, acknowledging the successful application of this algorithm for Outlet-TP modeling. In contrast, the complex structure of LSTM made it time-consuming and unstable coupled with noise, while other shallower architectures, i.e., RF, SVM, GTB, and ANFIS were unable to address large datasets with nonlinear and nonstationary behavior. Consequently, this study provides a reliable and accurate approach to forecast wastewater effluent quality, which is pivotal in terms of the socio-economic aspects of wastewater management.

Multi-step and Multi-Task Learning to Predict Quality-Related Variables in Wastewater Treatment Processes

Learning a neural network-based soft sensor with double-errors parallel optimization towards effluent variable prediction in wastewater treatment plants

Modeling of Adaptive Multi-Output Soft-Sensors with Applications in Wastewater Treatments

A Deep Semi-Supervised Learning Framework Towards Multi-Output Soft Sensors Development and Applications in Wastewater Treatment Processes

Process Monitoring of Quality-Related Variables in Wastewater Treatment Using Kalman-Elman Neural Network-Based Soft-Sensor Modeling

Interval Multiple-Output Soft Sensors Development with Capacity Control for Wastewater Treatment Applications: A Comparative Study

A novel two-step adaptive multioutput semisupervised soft sensor with applications in wastewater treatment

Exploring potential machine learning application based on big data for prediction of wastewater quality from different full-scale wastewater treatment plants

Modeling an Adaptive Hybrid Soft Sensor with Co-training Learning Toward Applications in Wastewater Treatment

A Hierarchical Soft-Sensor Using Spatiotemporal Information Transformation and ARMA with Application in Wastewater Treatment.

A hybrid deep learning approach to improve real-time effluent quality prediction in wastewater treatment plant

Quantitative Modeling and Predictive Analysis of Chemical Oxygen Demand in Wastewater Treatment Systems Utilizing Long Short-Term Memory Neural Network

Development of Semi-Supervised Multiple-Output Soft-Sensors with Co-training and Tri-Training MPLS and MRVM

Learning Soft Sensors Using Time Difference–based Multi-Kernel Relevance Vector Machine with Applications for Quality-Relevant Monitoring in Wastewater Treatment

Date-Driven Soft-Sensor Design for Biological Wastewater Treatment Using Deep Neural Networks and Genetic Algorithms

Real-time Water Quality Prediction of Wastewater Treatment Plants Using Advanced Deep Learning Networks

Water Quality Prediction Based on Multi-Task Learning

A Multi-Step Sequence-to-Sequence Model with Attention LSTM Neural Networks for Industrial Soft Sensor Application

Development of Multiple-Step Soft-Sensors Using a Gaussian Process Model with Application for Fault Prognosis

Brain-inspired multimodal approach for effluent quality prediction using wastewater surface images and water quality data

Learning Adaptive Semi-Supervised Multi-Output Soft-Sensors with Co-Training of Heterogeneous Models