# Prediction of Sludge Volume Index in a Wastewater Treatment Plant Using Recurrent Neural Network

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

#### 1.1. Sludge Volume Index (SVI)

#### 1.2. Activated Sludge Process

#### 1.3. Filamentous Bulking

#### 1.4. Recurrent Neural Network (RNN)

_{1}, …, x

_{t}), a standard RNN computes a series of outputs (y

_{1}, …, y

_{t}) by iterating the following equation [17]:

_{t}and output h

_{t}. A loop allows information to persist and pass from one step of the network to the next, in which traditional neural networks cannot handle this.

#### 1.5. Shortcomings of Previous Predictive Models

#### 1.6. Explainable Artificial Intelligence

## 2. Materials and Methods

#### 2.1. Data Collection

_{5}), Total Suspended Solids (TSS), Total Kjeldahl Nitrogen (TKN), Ammoniacal nitrogen (NH

_{3}N), Total Phosphorus (TP), and organic loading (flow rate × influent BOD

_{5}) were selected as inputs of the model, and the output is SVI. Then, data visualization was applied using the Python program. Figure 4 displays Sludge Volume Index (SVI) data from 1996 to 2020.

#### 2.2. Recurrent Neural Networks Models and Shapley Explanation

## 3. Results

_{5}, and flow rate are the most impact input parameters to the SVI prediction, followed by TP, TKN, TSS, and NH

_{3}N. Figure 17 shows that when SVI is 114.6, organic loading and flow rate lowered the predicted SVI value, and TP, TSS, NH

_{3}N, TKN, and BOD

_{5}increased the SVI value. Lastly, the explainable function can help determine input parameters that affect each output value in Figure 18. Lastly, Figure 19 shows the accuracy of the model using Mean Absolute Error (MAE) for training and testing set. The figure shows the lines are very low and close to each other, which means the model has a good performance. The most important finding is that the organic loading and TP in mass (concentration × flow rate) affect the SVI value most, implying that the WWTP might not be able to supply a proper amount of oxygen in response to the condition change. Thus, the real-time aeration control is thought to achieve a stable SVI.

_{5}, and flow rate were the most related parameters to SVI prediction, followed by TKN, TP, NH

_{3}N, and TSS. Depending on the operation condition, the principal parameters affecting the prediction varied. Therefore, applying this explainable function along with model prediction would assist the WWTP operation by closely monitoring the system, visualizing and controlling the system, making a model prediction, interpreting the result, and providing a faulty alarm.

## 4. Discussion

_{5}, and flow rate most affected SVI prediction. It can be caused by oxygen control in an aeration system because aeration impacts BOD

_{5}and SVI. Therefore, aeration control should be thoroughly monitored in an aeration system. Although it was possible to determine which parameter(s) caused higher SVI, reasons and corrective measures must be investigated further for individual WWTPs. The developed method can be applied to other WWTPs, but causative reasons may differ depending on the treatment process, characteristics of raw wastewater, air supply system, DO setpoint and control method, etc., suggesting different solutions for higher SVI.

## 5. Conclusions

_{5}and flow rate. Therefore, it is recommended to improve the aeration control system.

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Conflicts of Interest

## References

- Najy, M.; Talbi, F.Z.; Ech-Chafay, H.; Lachhab, M.; El Qryefy, M.; Belghyti, D. Exploitation of the domestic Wastewater Treatment Plant by activated sludge in the airport area of the city ben slimane (Morocco). Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci.
**2020**, XLIV-4/W3-2020, 303–308. [Google Scholar] [CrossRef] - Wikipedia Sludge Volume Index. Available online: https://en.wikipedia.org/w/index.php?title=Sludge_volume_index&oldid=963975303 (accessed on 8 July 2021).
- Djeddou, M.; Achour, B. The Use of a Neural Network Technique for the Prediction of Sludge Volume Index in Municipal Wastewater Treatment Plant. Larhyss J.
**2015**, 24, 351–370. [Google Scholar] - Ergun, H.; Özbay, Y.; Güçlü, D.; Küçükhemek, M. Prediction of Sludge Volume Index Bulking Using Image Analysis and Neural Network at a Full-Scale Activated Sludge Plant. Desalination Water Treat.
**2015**, 57, 17195–17205. [Google Scholar] [CrossRef] - Lou, I.; Zhao, Y. Sludge Bulking Prediction Using Principle Component Regression and Artificial Neural Network. Math. Probl. Eng.
**2012**, 2012, e237693. [Google Scholar] [CrossRef] [Green Version] - Richard, M. Activated Sludge Microbiology Problems and Their Control. In Proceedings of the the 20th Annual USEPA National Operator Trainers Conference, Buffalo, NY, USA, 8 June 2003; Volume 21. [Google Scholar]
- Phuc, B.D.H.; You, S.-S.; Hung, B.M.; Kim, H.-S. Robust Control Synthesis for the Activated Sludge Process. Environ. Sci. Water Res. Technol.
**2018**, 4, 992–1001. [Google Scholar] [CrossRef] - Yousuf, I. Methods for estimation and comparison of activated sludge settleability. In Proceedings of the 38th Annual WIOA Qld Water Industry Operations Conference, Parklands, Gold Coast, QLD, Australia, 4–6 June 2013; Volume 7. [Google Scholar]
- Edwards, J. Predictive Analytics: Transforming Data into Future Insights. Available online: https://www.cio.com/article/3273114/what-is-predictive-analytics-transforming-data-into-future-insights.html (accessed on 3 November 2021).
- DiPietro, R.; Hager, G.D. Chapter 21-Deep learning: RNNs and LSTM. In Handbook of Medical Image Computing and Computer Assisted Intervention; Zhou, S.K., Rueckert, D., Fichtinger, G., Eds.; The Elsevier and MICCAI Society Book Series; Academic Press: Cambridge, MA, USA, 2020; pp. 503–519. ISBN 978-0-12-816176-0. [Google Scholar]
- Qiao, J.F.; Yang, W.W.; Yuan, M.Z. Recurrent High Order Neural Network Modeling for Wastewater Treatment Process. J. Comput.
**2011**, 6, 1570–1577. [Google Scholar] [CrossRef] - Sajee, A. Model Complexity, Accuracy and Interpretability. Available online: https://towardsdatascience.com/model-complexity-accuracy-and-interpretability-59888e69ab3d (accessed on 3 November 2021).
- Bye, C.; Dold, P. Sludge Volume Index Settleability Measures: Effect of Solids Characteristics and Test Parameters. Water Environ. Res.
**1998**, 70, 87–93. [Google Scholar] [CrossRef] - Tesh, K. How to calculate sludge volume index-SVI. Water Wastewater Courses. Available online: https://www.waterandwastewatercourses.com/calculate-sludge-volume-index-svi/ (accessed on 11 November 2021).
- Finnegan, T. Operation and Control; Environmental Leverage, Inc.: Batavia, IL, USA, 2020; Volume 12. [Google Scholar]
- Jenkins, D.; Richard, M.G.; Daigger, G.T. Manual on the Causes and Control of Activated Sludge Bulking, Foaming, and Other Solids Separation Problems, 3rd ed.; CRC Press: Boca Raton, FL, USA, 2003; ISBN 978-1-56670-647-6. [Google Scholar]
- Sutskever, I.; Vinyals, O.; Le, Q.V. Sequence to Sequence Learning with Neural Networks. arXiv
**2014**, arXiv:14093215. [Google Scholar] - Olah, C. Understanding LSTM Networks-Colah’s Blog. Available online: https://colah.github.io/posts/2015-08-Understanding-LSTMs/ (accessed on 28 May 2020).
- Henze, M.; Gujer, W.; Mino, T.; van Loosedrecht, M. Activated Sludge Models ASM1, ASM2, ASM2d and ASM3. Water Intell. Online
**2015**, 5, 9781780402369. [Google Scholar] [CrossRef] - Harrou, F.; Dairi, A.; Sun, Y.; Senouci, M. Statistical Monitoring of a Wastewater Treatment Plant: A Case Study. J. Environ. Manag.
**2018**, 223, 807–814. [Google Scholar] [CrossRef] [PubMed] - Dochain, D.; Vanrolleghem, P.A. Dynamical Modelling & Estimation in Wastewater Treatment Processes; IWA publishing: London, UK, 2021; Volume 4, p. 9781780403045. [Google Scholar] [CrossRef]
- Ghorbani, A.; Abid, A.; Zou, J. Interpretation of Neural Networks Is Fragile. Proc. AAAI Conf. Artif. Intell.
**2019**, 33, 3681–3688. [Google Scholar] [CrossRef] [Green Version] - Choi, E.; Bahadori, M.T.; Kulas, J.A.; Schuetz, A.; Stewart, W.F.; Sun, J. RETAIN: An Interpretable Predictive Model for Healthcare Using Reverse Time Attention Mechanism. arXiv
**2017**, arXiv:160805745. [Google Scholar] - Lundberg, S.; Lee, S.-I. A Unified Approach to Interpreting Model Predictions. arXiv
**2017**, arXiv:170507874. [Google Scholar]

**Figure 10.**This is a figure of normal distribution. (

**a**) Normal distribution of the first dataset from 1996 to 2020; (

**b**) Normal distribution of the second dataset from 2001 to 2020; (

**c**) Normal distribution of the second dataset from 2010 to 2020.

**Figure 11.**This is a figure of a normal probability plot. (

**a**) Normal probability plot of the first dataset from 1996 to 2020; (

**b**) Normal probability plot of the second dataset from 2001 to 2020; (

**c**) Normal probability plot of the third dataset from 2010 to 2020.

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 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://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Wongburi, P.; Park, J.K.
Prediction of Sludge Volume Index in a Wastewater Treatment Plant Using Recurrent Neural Network. *Sustainability* **2022**, *14*, 6276.
https://doi.org/10.3390/su14106276

**AMA Style**

Wongburi P, Park JK.
Prediction of Sludge Volume Index in a Wastewater Treatment Plant Using Recurrent Neural Network. *Sustainability*. 2022; 14(10):6276.
https://doi.org/10.3390/su14106276

**Chicago/Turabian Style**

Wongburi, Praewa, and Jae K. Park.
2022. "Prediction of Sludge Volume Index in a Wastewater Treatment Plant Using Recurrent Neural Network" *Sustainability* 14, no. 10: 6276.
https://doi.org/10.3390/su14106276