What question did this study set out to answer?

This research aims to develop a federated machine learning framework to predict energy consumption in wastewater treatment plants using real-world data.

February 22, 2026International Journal of Environmental Research0 citationsOpen Access

Developing a Vertical Federated Machine Learning Framework for Predicting Energy Consumption in Wastewater Treatment Plants: A Case Study of Melbourne

BDBabür DeliktaşBursa Uludağ Üni̇versi̇tesi̇ATAiman TariqBursa Uludağ Üni̇versi̇tesi̇GKGökçen KurtBursa Uludağ Üni̇versi̇tesi̇

Key Points

This research aims to develop a federated machine learning framework to predict energy consumption in wastewater treatment plants using real-world data.
Developed a vertical federated machine learning framework (VFed-ML) for data integration.
Assessed various machine learning models, including regression and artificial neural networks.
Implemented feature selection techniques with XGBoost and Random Forest.
Analyzed training and validation loss curves to evaluate model performance.
The federated model effectively learned over training rounds, showing a decline in loss values.
Demonstrated improved predictive accuracy with minimal overfitting signs.
Identified key challenges, including data heterogeneity and communication bottlenecks, with recommendations for addressing them.

Abstract

This study presents a Vertical Federated Machine Learning (VFed-ML) framework designed to predict energy consumption in wastewater treatment plants (WWTPs) using actual data from a Melbourne-based facility. The effectiveness of various machine learning models including regression, ensemble learning, decision trees, and artificial neural networks were assessed for energy consumption prediction. Feature selection techniques using XGBoost and Random Forest were conducted to identify key variables that improve model performance. The proposed VFed-ML framework enables multiple agencies to collaboratively develop a unified machine learning model while ensuring the secure and efficient integration of multi-source data. The performance of the federated model was assessed over successive training rounds by analyzing the convergence of training and validation loss curves. Overall, the progressive decline in loss values during the later training rounds indicates that the model is effectively learning and improving its performance over time. The final model demonstrated improved predictive accuracy with minimal signs of overfitting, highlighting its ability to generalize well. These findings suggest that the proposed approach can be adapted to other WWTPs by incorporating plant-specific data. Key challenges such as data heterogeneity, communication bottlenecks, and model convergence, were identified and recommendations to address these issues were provided. The study also outlines opportunities for future research, including the incorporation of additional variables related to operational and external factors. In conclusion, the VFed-ML framework offers a promising solution for optimizing energy consumption predictions in WWTPs, safeguarding data privacy and security. Its flexibility allows for customization to other plants, highlighting its potential for real-world applications in environmental monitoring and resource management.

AI से पूछें

Bookmark

View Full Paper