Data-Driven Approaches for Quantitative and Qualitative Control of Urban Drainage Systems (Preliminary Results)

Abstract

The uncontrolled urbanization of soil leads to two main effects: the increase in flood discharges due to changes in permeability capacity and the negative impact in terms of quality on water bodies. These effects can be mitigated by common engineering practices, such as Low Impact Development (LID, which generally involves stormwater treatment devices on a smaller scale rather than centralized solutions); Sustainable Urban Drainage Systems (SUDSs, a range of technologies and techniques used to drain stormwater in a more sustainable manner than conventional solutions); Best Management Practices (BMPs, suggested solutions are more focalized on pollution prevention in urban systems), and more. Among the proposed solutions, detention/retention systems and stormwater ponds can also perform excellent functions with regard to hydraulic hazards and both quantitative and qualitative control of sewer discharge, thanks to stormwater volume accumulation together with the presence of vegetation, when the basin is conceived as a natural-looking lake environment. The use of data-driven approaches could represent an effective approach for the prediction of the characteristics of the sewage tributaries, for the generation of synthetic time series of quantitative/qualitative data of sewer flows or for Real-Time Control (RTC) to reduce overflow at the Waste Water Treatment Plant (WWTP). This work shows the preliminary results obtained by applying NARX neural networks in order to estimate quality indices (the turbidity in this study) in sewer systems. The available data are discharge, temperature, gage height, specific conductivity, and precipitation, whose use as parameters for the recurrent neural network leads to values of R = 0.77–0.80 in the various combinations tested.

Keywords

• artificial neural networks
• NARX
• stormwater quality
• urban watersheds
• machine learning algorithms
• first flush