[Paper Review] DeepCSO: Forecasting of Combined Sewer Overflow at a Citywide Level using Multi-task Deep Learning

Duo Zhang, Geir Lindholm+1 more|arXiv (Cornell University)|Nov 9, 2018

Flood Risk Assessment and ManagementEnvironmental Science15 references3 citations

TL;DR

DeepCSO proposes a multi-task deep learning model to forecast combined sewer overflow (CSO) events across an entire city's sewer network in near real time. By integrating data-driven flexibility with structural insights from deterministic models, the approach outperforms traditional methods in accuracy and scalability for citywide CSO prediction.

ABSTRACT

Combined Sewer Overflow (CSO) is a major problem to be addressed by many cities. Understanding the behavior of sewer system through proper urban hydrological models is an effective method of enhancing sewer system management. Conventional deterministic methods, which heavily rely on physical principles, is inappropriate for real-time purpose due to their expensive computation. On the other hand, data-driven methods have gained huge interests, but most studies only focus on modeling a single component of the sewer system and supply information at a very abstract level. In this paper, we proposed the DeepCSO model, which aims at forecasting CSO events from multiple CSO structures simultaneously in near real time at a citywide level. The proposed model provided an intermediate methodology that combines the flexibility of data-driven methods and the rich information contained in deterministic methods while avoiding the drawbacks of these two methods. A comparison of the results demonstrated that the deep learning based multi-task model is superior to the traditional methods.

Motivation & Objective

• Address the limitations of conventional deterministic hydrological models, which are computationally expensive and unsuitable for real-time applications.
• Overcome the shortcomings of existing data-driven methods that focus on isolated sewer components and provide only abstract-level insights.
• Develop a scalable, citywide forecasting system that captures complex interactions across multiple CSO structures simultaneously.
• Bridge the gap between data-driven models and physics-based models by incorporating structural information while maintaining computational efficiency.
• Enable near real-time prediction of CSO events to support proactive urban drainage management and flood mitigation.

Proposed method

• Design a multi-task deep learning framework that jointly predicts overflow events at multiple CSO outfalls across a citywide sewer network.
• Use long short-term memory (LSTM) networks to model temporal dependencies in rainfall and flow data across multiple locations.
• Integrate structural and spatial information from the sewer network topology to guide the learning process and improve generalization.
• Train the model end-to-end using historical rainfall and flow data from multiple CSO monitoring points to capture inter-location dynamics.
• Apply shared representation learning across tasks to improve generalization and reduce overfitting on limited data.
• Optimize the model using a multi-task loss function that balances prediction accuracy across all CSO locations.

Experimental results

Research questions

• RQ1 Can a deep learning model effectively forecast CSO events at multiple locations across a city simultaneously with high accuracy?
• RQ2 How does the integration of structural sewer network information improve the performance of data-driven CSO forecasting models?
• RQ3 To what extent does multi-task learning enhance prediction accuracy and generalization compared to single-task models?
• RQ4 Can the proposed model achieve real-time forecasting capabilities while maintaining high predictive performance?
• RQ5 How does the model compare to traditional deterministic and single-task data-driven approaches in terms of accuracy and scalability?

Key findings

• The DeepCSO model significantly outperforms traditional deterministic models in forecasting accuracy, particularly in capturing complex, nonlinear dynamics across the sewer network.
• Multi-task learning improves prediction performance across all CSO locations by enabling knowledge transfer between related tasks.
• The integration of sewer network topology enhances model generalization and reduces overfitting, especially in data-scarce locations.
• The model achieves near real-time inference capability, making it suitable for operational use in urban drainage management systems.
• Quantitative results demonstrate superior performance in terms of mean absolute error (MAE) and coefficient of determination (R²) compared to baseline models.
• The model effectively captures spatial and temporal dependencies across the citywide sewer network, enabling coordinated flood mitigation strategies.