In the face of climate change and evolving environmental regulations, effective management and forecasting of sewage water systems (SWS) under various scenarios have become important. These systems wield significant influence over urban flood control and water quality treatments. In this context, a Data-Driven Digital Twin (DT) is developed specifically for a small SWS basin located in northern Italy. This basin encompasses a sewage network, featuring Doppler sensors that measure water velocity, pressure (depth), and temperature every six minutes. Additionally, rain gauges provide minute-by-minute data. Given the operational conditions of these sensors, occasional low-quality measurements are inevitable. To address this issue, a Neural Network (NN) is designed, trained, and integrated into the DT capable of identifying anomalous values, attributing potential causes (e.g., sensor contamination), and suggesting accurate replacements. This research explores regressive neural network models approaches: a convolutional layer neural network (CNN), CNN with Long Short-Term Memory (LSTM) approach, and a CNN model with residual connection. Models are all replicate the SWS configuration and share the same training data. These models are rigorously evaluated under scenarios involving missing data, such as sensor removal, and both consistently exhibit a high general accuracy rate exceeding 90%. This project not only showcases the successful development and application of the DT but also underscores the importance of collaboration between industry, government, and academia in addressing critical environmental challenges.
In the face of climate change and evolving environmental regulations, effective management and forecasting of sewage water systems (SWS) under various scenarios have become important. These systems wield significant influence over urban flood control and water quality treatments. In this context, a Data-Driven Digital Twin (DT) is developed specifically for a small SWS basin located in northern Italy. This basin encompasses a sewage network, featuring Doppler sensors that measure water velocity, pressure (depth), and temperature every six minutes. Additionally, rain gauges provide minute-by-minute data. Given the operational conditions of these sensors, occasional low-quality measurements are inevitable. To address this issue, a Neural Network (NN) is designed, trained, and integrated into the DT capable of identifying anomalous values, attributing potential causes (e.g., sensor contamination), and suggesting accurate replacements. This research explores regressive neural network models approaches: a convolutional layer neural network (CNN), CNN with Long Short-Term Memory (LSTM) approach, and a CNN model with residual connection. Models are all replicate the SWS configuration and share the same training data. These models are rigorously evaluated under scenarios involving missing data, such as sensor removal, and both consistently exhibit a high general accuracy rate exceeding 90%. This project not only showcases the successful development and application of the DT but also underscores the importance of collaboration between industry, government, and academia in addressing critical environmental challenges.
Calibration and Comparison of Regressive Neural Network Models for Environmental Parameter Forecasting for Sewage Water System
DAVUT, EMRE
2022/2023
Abstract
In the face of climate change and evolving environmental regulations, effective management and forecasting of sewage water systems (SWS) under various scenarios have become important. These systems wield significant influence over urban flood control and water quality treatments. In this context, a Data-Driven Digital Twin (DT) is developed specifically for a small SWS basin located in northern Italy. This basin encompasses a sewage network, featuring Doppler sensors that measure water velocity, pressure (depth), and temperature every six minutes. Additionally, rain gauges provide minute-by-minute data. Given the operational conditions of these sensors, occasional low-quality measurements are inevitable. To address this issue, a Neural Network (NN) is designed, trained, and integrated into the DT capable of identifying anomalous values, attributing potential causes (e.g., sensor contamination), and suggesting accurate replacements. This research explores regressive neural network models approaches: a convolutional layer neural network (CNN), CNN with Long Short-Term Memory (LSTM) approach, and a CNN model with residual connection. Models are all replicate the SWS configuration and share the same training data. These models are rigorously evaluated under scenarios involving missing data, such as sensor removal, and both consistently exhibit a high general accuracy rate exceeding 90%. This project not only showcases the successful development and application of the DT but also underscores the importance of collaboration between industry, government, and academia in addressing critical environmental challenges.File | Dimensione | Formato | |
---|---|---|---|
Davut_Emre.pdf
accesso aperto
Dimensione
13.99 MB
Formato
Adobe PDF
|
13.99 MB | Adobe PDF | Visualizza/Apri |
The text of this website © Università degli studi di Padova. Full Text are published under a non-exclusive license. Metadata are under a CC0 License
https://hdl.handle.net/20.500.12608/60579