Applying Recurrent Neural Networks and Blocked Cross-Validation to Model Conventional Drinking Water Treatment Processes - Article

Aug 4, 2026

Upload by Violet

The jar test serves as the standard approach for predicting conventional drinking water treatment performance and optimizing coagulant dose, but its time-consuming nature and reliance on human intervention limit continuous predictions. A machine learning model is developed from historical DWT plant data to run continuously with real-time sensor inputs, forecasting clarified water turbidity 15 minutes ahead. Three model families—multilayer perceptron, LSTM recurrent neural network, and GRU recurrent neural network—are compared under two training strategies: holdout and blocked cross-validation. Results show GRU-based RNN as the best overall option, reaching a mean absolute error as low as 0.044 NTU on an independent production set, while BCV-trained models typically match or outperform holdout-trained counterparts.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Introduction

Drinking water treatment unit processes and modeling need

Methods

ML models for turbidity forecasting (MLP, LSTM, GRU)

Training strategies (holdout vs blocked cross-validation)

Results

Model comparison on independent production data

Error analysis and impact of blocked cross-validation

Conclusions

Key findings and implications for DWT process modeling

Titre: Applying recurrent neural networks and blocked cross-validation to Title: model conventional drinking water treatment processes Auteurs: Aleksandar Jakovljevic, Laurent Charlin, & Benoit Barbeau Authors: Date: 2024 Type: Article de revue / Article Jakovljevic, A. , Charlin, L. , & Barbeau, B. (2024) . Applying recurrent neural Référence: networks and blocked cross-validation to model conventional drinking water Citation: treatment processes. Water, 16(7), 16071042 (14 pages) . [https://doi.org/10.3390/w16071042](https://doi.org/10.3390/w16071042) | Document en libre accès dans PolyPublie
Open Access document in PolyPublie
| | | | --- | --- | --- | | URL de PolyPublie:
PolyPublie URL: | [https://publications.polymtl.ca/58097/](https://publications.polymtl.ca/58097/) | | | Version officielle de l'éditeur / Published version
Version:
Révisé par les pairs / Refereed
| | | | Conditions d’utilisation:
CC BY
Terms of Use: | | | | Document publié chez l’éditeur officiel
Document issued by the official publisher
| | | | | | | --- | --- | --- | --- | --- | --- | | Titre de la revue:
Water (vol. 16, no. 7)
Journal Title: | | | | | | | Maison d’édition:
Publisher: | | | MDPI | | | | | | URL officiel:
Official URL:
| | | | | | Mention légale:
Legal notice: | | © 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license ([htt](https://creativecommons.org/licenses/by/4.0/)[ps:](https://creativecommons.org/licenses/by/4.0/)[//crea](https://creativecommons.org/licenses/by/4.0/)[tiv](https://creativecommons.org/licenses/by/4.0/)[ecom](https://creativecommons.org/licenses/by/4.0/)[m](https://creativecommons.org/licenses/by/4.0/)[ons.org/licenses/b](https://creativecommons.org/licenses/by/4.0/)[y/4](https://creativecommons.org/licenses/by/4.0/)[.0/](https://creativecommons.org/licenses/by/4.0/)). | | | Ce fichier a été téléchargé à partir de PolyPublie, le dépôt institutionnel de Polytechnique Montréal This file has been downloaded from PolyPublie, the institutional repository of Polytechnique Montréal [https://publications.polymtl.ca](https://publications.polymtl.ca) water Article Applying Recurrent Neural Networks and Blocked Cross-Validation to Model Conventional Drinking Water Treatment Processes Aleksandar Jakovljevic 1,*, Laurent Charlin 2 and Benoit Barbeau 1 Citation: Jakovljevic, A.; Charlin, L.; Barbeau, B. Applying Recurrent Neural Networks and Blocked Cross-Validation to Model Conventional Drinking Water Treatment Processes. Water 2024, 16, 1042. [https://doi.org/10.3390/](https://doi.org/10.3390/)[ ](https://doi.org/10.3390/)w16071042 Academic Editor: Long Ho Received: 8 February 2024 Revised: 27 March 2024 Accepted: 28 March 2024 Published: 4 April 2024 Copyright: © 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license ([https://](https://)[ ](https://)[creativecommons.org/licenses/by/](creativecommons.org/licenses/by/)[ ](creativecommons.org/licenses/by/)[4.0/](4.0/)) . 1 Department of Civil, Geological and Mining Engineering, Polytechnique Montréal, 2500 Chemin de Polytechnique, Montréal, QC H3T 1J4, Canada; [benoit.barbeau@polymtl.ca](benoit.barbeau@polymtl.ca) 2 Department of Decision Sciences, HEC Montréal, 3000 Chemin de la Côte-Sainte-Catherine, Montréal, QC H3T 2A7, Canada; [laurent.charlin@hec.ca](laurent.charlin@hec.ca) * Correspondence: [aleksandar.jakovljevic@polymtl.ca](aleksandar.jakovljevic@polymtl.ca) Abstract: The jar test is the current standard method for predicting the performance of a conventional drinking water treatment (DWT) process and optimizing the coagulant dose. This

FAQs

Why is the jar test not suitable for continuous drinking water treatment predictions?

The jar test is time-consuming and requires human intervention, making it impractical for continuous process forecasting.

What is predicted and how far ahead in time?

The model predicts clarified water turbidity 15 minutes in advance using real-time sensor data.

Which model performed best and what was the reported error?

A GRU-based RNN performed best overall, achieving a mean absolute error as low as 0.044 NTU on an independent production set.

Aug 4, 2026

Upload by Violet