Data-driven Modeling of Combined Sewer Systems for Urban Sustainability: An Empirical Evaluation

TL;DR

This work tackles predicting Combined Sewer System (CSS) overflow basin water levels under extreme rainfall using data-driven time-series models. It conducts a comprehensive empirical evaluation of six neural architectures (including LSTM and TFT) under two modeling paradigms: global models that fuse exogenous data and local models that rely solely on historical sensor readings. Using three years of real-world maintenance data from Duisburg, the study shows global models generally yield higher accuracy, with LSTM and TFT delivering the best median performance, while local models remain viable for outage situations and edge deployment due to lower resource demands. The findings advance urban sustainability by enabling more reliable load balancing in CSS, reducing overflow risk, and informing resilient, cost-effective management strategies for utility operators. The work also highlights avenues for future work, including edge deployment and potential physics-informed hybrids to further enhance predictive robustness.

Abstract

Climate change poses complex challenges, with extreme weather events becoming increasingly frequent and difficult to model. Examples include the dynamics of Combined Sewer Systems (CSS). Overburdened CSS during heavy rainfall will overflow untreated wastewater into surface water bodies. Classical approaches to modeling the impact of extreme rainfall events rely on physical simulations, which are particularly challenging to create for large urban infrastructures. Deep Learning (DL) models offer a cost-effective alternative for modeling the complex dynamics of sewer systems. In this study, we present a comprehensive empirical evaluation of several state-of-the-art DL time series models for predicting sewer system dynamics in a large urban infrastructure, utilizing three years of measurement data. We especially investigate the potential of DL models to maintain predictive precision during network outages by comparing global models, which have access to all variables within the sewer system, and local models, which are limited to data from a restricted set of local sensors. Our findings demonstrate that DL models can accurately predict the dynamics of sewer system load, even under network outage conditions. These results suggest that DL models can effectively aid in balancing the load redistribution in CSS, thereby enhancing the sustainability and resilience of urban infrastructures.

Figures (4)

• Figure 1: CSS with overflow basins collect rainwater and wastewater into multiple basins. During heavy rainfall, these basins may exceed capacity, leading to the untreated mixture overflow into the environment chiaburu2024interpretable.
• Figure 2: Global models often outperform the local models. The distribution of test MSE (upper) and MAPE (lower) across model types and approaches, based on multiple training runs with different random parameter initializations, show that the lowest median MSE values were obtained with LSTM and TFT models. The lowest median MAPE values were obtained with even simpler models such as TCN.
• Figure 3: The global LSTM model demonstrates better forecasting performance than the local TFT model. Forecasts from the global LSTM model (see Figure \ref{['fig:lstm_single1']} and \ref{['fig:lstm_single2']}) and local TFT model (see Figure \ref{['fig:tft_single1']} and \ref{['fig:tft_single2']}) were evaluated on two samples. The plots show the forecasts for the water level of the overflow basin. The dashed red line indicates the start of the forecast window. Samples before the dashed red line represent the input data for the model, spanning 72 hours, while the model produces a 12-hour forecast, shown in blue.
• Figure 4: Global models are able to predict sudden changes much better than local models. 12-Hour Ahead Forecasts of the global LSTM (left) and local TFT (right) of Filling Levels Throughout 2023.