Toward smart wastewater treatment plants: a novel data-driven sludge blanket model based on stochastic differential equations

Abstract

A novel data-driven model for forecasting the sludge blanket height in secondary clarifiers is presented. The model is trained on sensor measurements of the sludge blanket height and used as inputs such as (1) the clarifier feed flow rate, (2) feed suspended solids concentration, and (3) the clarifier recycle flow rate. The model’s prediction accuracy is evaluated based on data from two Danish wastewater treatment plants by means of root-mean-square errors (RMSEs), and results are compared against a persistence model. We demonstrate that the developed model is superior to the persistence forecast at both plants during high blanket dynamics. In the best scenario, the model improves the RMSE by 0.1/0.4 m at prediction horizons of 2.5/10 h, assuming known inputs. The model performance is subsequently considered with forecasted inputs using two different forecast scenarios. We discuss differences in the two plants’ performance and requirements to achieve good model performance. The model is well-suited for a model predictive control strategy, whose purpose ultimately is to improve clarifier control, increasing hydraulic capacity and reducing overflow suspended solids.