Author:

Keywords:

Artifical Neural Network (ANN), Black-box Model, Combined Sewer Overflows (CSOs), WasteWater Treatment Plant (WWTP) influent

Abstract:

When building integrated river water quantity-quality models it is paramount that also the pollution originating for Combined Sewer Overflows (CSOs) can be taken into account. However, even though it is known that CSO water is polluted, little is yet known about the magnitude of this pollution. We hypothesize that the WWTP influent measurements can serve as a proxy of the water quality at the CSO if not located too far from each other. Many models are already available to simulate the dynamics of the pollutant concentrations at the WWTP influent. Most of these models, however, require a continuous series of historical water quality measurements at high temporal resolution (order of magnitude of minutes). In this study, a Multi-Layer Perceptron neural network (MLP NN) is applied to generate daily concentration time series by training to measurements available at a weekly resolution. Temporal interpolation is based on daily discharge data and hourly rainfall intensities. This approach is tested to simulate ammonia (NH4), orthophosphate (OP), and total phosphorus (TP) for 15 WWTPs in Flanders, Belgium, showing promising results.