Publicação
A machine learning approach to forecast influent flow in wastewater treatment plants
| Resumo: | Influent flow prediction is a crucial task in Wastewater Treatment Plant (WWTP) operations, as it enables the optimization of treatment processes and ensures efficient resource utilization. Without accurate flow prediction, WWTPs may experience operational inefficiencies, leading to increased energy consumption and potential treatment issues. As environmental concerns grow, the need for precise flow prediction becomes increasingly critical for sustainable wastewater management. The integration of Deep Learning (DL) techniques in WWTP operations can show a significant impact in these infrastructures due to their ability to enhance inflow prediction and capture complex temporal patterns. Through these methods, it is possible to identify intricate relationships in time series data that traditional statistical approaches might miss, while also adapting to the non-linear nature of influent flow patterns. In this way, this dissertation aims to predict the influent flow for the next two days in a WWTP. Hence, several DL candidate models, such as Bidirectional Long Short-Term Memory (BiLSTM), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Multi-Layer Perceptron (MLP), are conceived, tuned and evaluated in multivariate and univariate scenarios. Through the analysis of the obtained results, namely through performance metrics, it was possible to verify that the BiLSTM and CNN candidate models generally outperform alternative architectures, with the univariate BiLSTM model achieving the minimal Root Mean Square Error (RMSE) of 243.01 3/, while the multivariate BiLSTM model recorded a RMSE of 250.713/. With this research, it is possible to leverage the discussion on the intersection of Machine Learning (ML) and wastewater engineering, providing a foundation for informed decision-making in the operation of WWTP. |
|---|---|
| Autores principais: | Fernandes, Pedro Gomes Machado Monteiro |
| Assunto: | Deep Learning Influent Flow Machine Learning Time Series Forecasting Wastewater Treatment Plants Aprendizagem Automática Aprendizagem Profunda Caudal Afluente Estações de Tratamento de Águas Residuais Previsão de Séries Temporais Engenharia e Tecnologia::Outras Engenharias e Tecnologias |
| Ano: | 2025 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Influent flow prediction is a crucial task in Wastewater Treatment Plant (WWTP) operations, as it enables the optimization of treatment processes and ensures efficient resource utilization. Without accurate flow prediction, WWTPs may experience operational inefficiencies, leading to increased energy consumption and potential treatment issues. As environmental concerns grow, the need for precise flow prediction becomes increasingly critical for sustainable wastewater management. The integration of Deep Learning (DL) techniques in WWTP operations can show a significant impact in these infrastructures due to their ability to enhance inflow prediction and capture complex temporal patterns. Through these methods, it is possible to identify intricate relationships in time series data that traditional statistical approaches might miss, while also adapting to the non-linear nature of influent flow patterns. In this way, this dissertation aims to predict the influent flow for the next two days in a WWTP. Hence, several DL candidate models, such as Bidirectional Long Short-Term Memory (BiLSTM), Convolutional Neural Network (CNN), Gated Recurrent Unit (GRU), Long Short-Term Memory (LSTM), and Multi-Layer Perceptron (MLP), are conceived, tuned and evaluated in multivariate and univariate scenarios. Through the analysis of the obtained results, namely through performance metrics, it was possible to verify that the BiLSTM and CNN candidate models generally outperform alternative architectures, with the univariate BiLSTM model achieving the minimal Root Mean Square Error (RMSE) of 243.01 3/, while the multivariate BiLSTM model recorded a RMSE of 250.713/. With this research, it is possible to leverage the discussion on the intersection of Machine Learning (ML) and wastewater engineering, providing a foundation for informed decision-making in the operation of WWTP. |
|---|
Atividades financiadas
Carregando projetos financiados...