Assessing the Performance of Deep Learning Algorithms With and Without Transfer Learning in Similar or Identical Wastewater Treatment Processes DOI Creative Commons

Jaeil Kim,

Sang-Ik Suh, Yongtae Ahn

et al.

Journal of Korean Society of Environmental Engineers, Journal Year: 2024, Volume and Issue: 46(3), P. 111 - 117

Published: March 31, 2024

This study assessed the feasibility of transfer learning from one wastewater treatment process to another using two popular deep algorithms. Specifically, convolutional neural network (CNN) and long short-term memory (LSTM), which consisted four three hidden layers, respectively, were used as benchmark algorithms for learning. Input data both provided plants with identical trains in series (located Jinju Cheongju City) over five-year period 2018 2022. Performance evaluation was also done not only against but those adopting strategies, freezing all layers developed pre-trained model other training last layer among multiple ones, respect Mean Squared Error (MSE). We found that performance CNN LSTM relatively comparative regardless dependent variables, discharge biochemical oxygen demand (BOD), whereas prediction accuracy slightly higher than BOD due its low variability. When froze existing applied algorithms, predictive improved discharge. Also, there no measurable variation approach. Potential applications include rapid reuse models (developed source domains) target domains are hard develop new lack

Language: Английский

Urban wastewater-based epidemiology for multi-viral pathogen surveillance in the Valencian region, Spain DOI Creative Commons
Inés Girón‐Guzmán, Enric Cuevas‐Ferrando, Regino Barranquero

et al.

Water Research, Journal Year: 2024, Volume and Issue: 255, P. 121463 - 121463

Published: March 16, 2024

Wastewater-based epidemiology (WBE) has lately arised as a promising tool for monitoring and tracking viral pathogens in communities. In this study, we analysed WBE's role multi-pathogen surveillance strategy to detect the presence of several illness causative agents. Thus, an epidemiological study was conducted from October 2021 February 2023 estimate weekly levels Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), Syncytial virus (RSV), Influenza A (IAV) influent wastewater samples (n = 69). parallel, one-year (October 2022) performed assess pathogenic human enteric viruses. Besides, proposed fecal contamination indicators crAssphage Pepper mild mottle (PMMoV) also assessed, along with plaque counting somatic coliphages. Genetic material rotavirus (RV), astrovirus (HAStV), norovirus genogroup I (GI) GII found almost all samples, while hepatitis E viruses (HAV HEV) only tested positive 3.77 % 22.64 respectively. No seasonal patterns were overall viruses, although RVs had peak prevalence winter months. All SARS-CoV-2 RNA, mean concentration 5.43 log genome copies per liter (log GC/L). The circulating variants concern (VOCs) by both duplex RT-qPCR next generation sequencing (NGS). Both techniques reliably showed how dominant VOC transitioned Delta Omicron during two weeks Spain December 2021. RSV IAV peaked months concentrations 6.40 4.10 GC/L, Moreover, three selected respiratory strongly correlated reported clinical data when normalised physico-chemical parameters presented weaker correlations normalising sewage or coliphages titers. Finally, predictive models generated each virus, confirming high reliability on WBE early-warning system communities system. Overall, presents optimal reflecting circulation diseases trends within area, its value stands out due public health interest.

Language: Английский

Citations

10
Impact of wastewater characteristics and weather events on the N2 and N1 gene target ratios during wastewater surveillance of SARS-CoV-2 at five treatment plants and an upper sewershed location DOI Creative Commons

Lena Carolin Bitter,

Richard Kibbee,

Tim Garant

et al.

The Science of The Total Environment, Journal Year: 2025, Volume and Issue: 981, P. 179592 - 179592

Published: May 9, 2025

Language: Английский

Citations

0
Process innovations and circular strategies for closing the water loop in a process industry DOI Creative Commons

Efthalia Karkou,

Athanasios Angelis-Dimakis,

Marco Parlapiano

et al.

Journal of Environmental Management, Journal Year: 2024, Volume and Issue: 370, P. 122748 - 122748

Published: Oct. 2, 2024

Language: Английский

Citations

1
Response time of fast flowing hydrologic pathways controls sediment hysteresis in a low-gradient watershed, as evidenced from tracer results and machine learning models DOI
Arlex Marin-Ramirez, David Tyler Mahoney,

Brenden Riddle

et al.

Journal of Hydrology, Journal Year: 2024, Volume and Issue: unknown, P. 132207 - 132207

Published: Oct. 1, 2024

Language: Английский

Citations

1
Assessing the Performance of Deep Learning Algorithms With and Without Transfer Learning in Similar or Identical Wastewater Treatment Processes DOI Creative Commons

Jaeil Kim,

Sang-Ik Suh, Yongtae Ahn

et al.

Journal of Korean Society of Environmental Engineers, Journal Year: 2024, Volume and Issue: 46(3), P. 111 - 117

Published: March 31, 2024

This study assessed the feasibility of transfer learning from one wastewater treatment process to another using two popular deep algorithms. Specifically, convolutional neural network (CNN) and long short-term memory (LSTM), which consisted four three hidden layers, respectively, were used as benchmark algorithms for learning. Input data both provided plants with identical trains in series (located Jinju Cheongju City) over five-year period 2018 2022. Performance evaluation was also done not only against but those adopting strategies, freezing all layers developed pre-trained model other training last layer among multiple ones, respect Mean Squared Error (MSE). We found that performance CNN LSTM relatively comparative regardless dependent variables, discharge biochemical oxygen demand (BOD), whereas prediction accuracy slightly higher than BOD due its low variability. When froze existing applied algorithms, predictive improved discharge. Also, there no measurable variation approach. Potential applications include rapid reuse models (developed source domains) target domains are hard develop new lack

Language: Английский

Citations

0