Prediction of influent composition in wastewater and sludge based on Statistical and Machine Learning models - Robotics and Control master thesis

Aug 5, 2026

Upload by Asher

Optimized operation of wastewater resource recovery facilities depends on managing disturbances such as variations in influent flow and wastewater compositions. Online monitoring of influent characteristics is limited by insufficient instrumentation and high costs. This thesis presents data-driven prediction of influent compositions for two wastewater treatment plants, using statistical time-series approaches and machine learning models trained on Hias wastewater and Veas sludge process data. Models including ARIMA, SARIMAX, regression variants, and ML algorithms are evaluated to estimate key inlet variables.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages95

Reading Time3 h 59 min

AccessFree to download

1. Introduction

1.1 Background

Research questions

1.2 Theoretical Background

1.2.1 ARIMA model

1.2.2 SARIMA Model

1.2.3 SARIMAX

ACIT5900 MASTER THESIS in Applied Computer and Information Technology (ACIT) May 2023 Robotics and Control Prediction of influent composition in wastewater and sludge based on Statistical and Machine Learning models. Bipasha Mukherjee S366272 Department of Mechanical, Electronics and Chemical Engineering Faculty of Technology, Art and Design Acknowledgement This project report is part of the course ACIT5900-1 21H Master thesis(short) Research. This report includes research on influent composition prediction of wastewater and sludge treatment process using statistical model for time series analysis and machine learning models. Being a student of Robotics and Control system, I can use my knowledge of both process control and automation in this thesis topic. This makes it easy to choose the thesis topic in this field. I want to express my sincere gratitude to my professors Tine Komulainen, Olga Korostynska, Arvind Keprate, Simen Antonsen, Rafael Borrajo, Per Ola Rønning, and Yasha Parvini for giving me valuable guidance through all master thesis discussion sessions. Regular weekly meeting sessions and presentations helped me to proceed further. They always helped me whenever or wherever I faced difficulties in the whole journey of master thesis. They also support in getting additional resources and facilities from OsloMet. I would like to thank Hias Water Resource Recovery Facility and VEAS Water Resource Recovery Facility for providing the datasets and helping to work with their process data. I would also like to thank OsloMet for giving me opportunity to study master in the university. I would also like to express my sincere gratitude to my classmates , Einar Nermo, Mohamed Abdishakur Mohamed, Bilal Mukhter to explain all my doubts through discussions, provide the support to complete the thesis assignments and thesis work . More personally, I want to thank my family for their patience, motivation, and support during my study and without whom this would not have been possible. Bipasha Mukherjee Oslo, Norway May 2023 Abstract For the optimized operation of a wastewater resource recycle facility (WWRF), it is essential to consider significant disturbances such as fluctuations in the influent flow rate and wastewater compositions. The online monitoring of influent characteristics is limited by scarce instrumentation and high costs. This study demonstrated influent composition prediction of two different wastewater treatment plants (WWTPs), with wastewater and sludge treatment process. Data-driven models (statistical models used for time series analysis/ Machine learning model) have been developed using HIAS wastewater treatment process and VEAS sludge treatment process data to predict the influent compositions. In this work, statistical models for time series analysis such as ARIMA (Autoregressive Integrated Moving Average) and SARIMAX (Seasonal Autoregressive Integrated Moving Average with Exogenous input), Linear regression, Lasso, Ridge regression and different machine learning algorithms such as Random Forest (RF), Decision Tree (DT), Support Vector Regression (SVR) and Artificial Neural Network (ANN) were examined and compared. These models were developed to detect inlet phosphate (PO4), and inlet soluble chemical oxygen demand(sCOD) in wastewater inlet organic acid in sludge, which served as output variables. In both processes, Linear regression, Ridge regression and Neural Network consistently demonstrated the best performance for evaluation estimation as evidenced by the lowest values of Root Mean Square Error (RMSE), and the highest coefficient of determination (R2) . SARIMAX exhibited acceptable results with R2 as 0.95 in organic acid prediction modeling. In contrast, ARIMA and SARIMAX algorithms in Hias datasets did not meet the requirements because of the complex and nonlinear structure of the dataset issue. This study offers an efficient method for forecasting the quality of wastew

FAQs

What problem does the thesis address?

The thesis targets the need to predict influent compositions in wastewater and sludge treatment processes where online monitoring is constrained by limited instrumentation and cost.

Which datasets and outputs are used in the study?

The models are trained on HIAS wastewater treatment data and VEAS sludge treatment data to predict outputs including inlet phosphate (PO4) and inlet soluble chemical oxygen demand (sCOD), as well as inlet organic acid variables in sludge.

How do the statistical and machine learning models perform?

Linear regression, ridge regression, and neural networks deliver the best overall performance with lowest RMSE and highest R2, while SARIMAX shows acceptable results (e.g., R2 ≈ 0.95 for organic acid prediction). ARIMA and SARIMAX under HIAS datasets do not meet requirements due to complex nonlinear dataset structure.

Aug 5, 2026

Upload by Asher