Modelling sewer and combined sewer overflows through a combination of linear reservoirs and neural networks

Changing precipitation extremes and more attention for water quality (e.g. the revised Urban Wastewater Treatment Directive in the EU) increase the need to understand, model and monitor sewer flow and especially combined sewer overflows (CSO’s), a major source of pollutants in urban areas.

The first step to model water quality correctly, is the correct modelling of water quantity. Additionally, to be able to test different setups and use these models in a complex modelling chain (e.g. in digital twins), there is a need for fast and correct models for the urban hydrological and sewer network.

Here, we present such a fast approach, allowing for a conceptual modelling of the urban sewer network. This is carried out by a combination of linear reservoirs, which resembles distinct zones within the urban area, and a neural network, which is applied to model the dry weather flow. By splitting the rain-driven and dry weather flow, the model can be more easily setup to correctly model sewer overflow, while simultaneously leveraging long-term area-specific relationships between the measured dry weather flow at the waste water treatment plant and the precipitation deficit.