Neural network to analyze wastewater treatment plant with cept

Overview Wastewater treatment is a complex process that is difficult to model and optimize. Operational data from the wastewater treatment plant Høvringen (HØRA) in Trondheim, Norway (a chemically enhanced primary treatment (CEPT) plant with coagulation, flocculation and particle separation by settling) has been used to train and test an artificial neural network (ANN) to predict the turbidity of the treated water given relevant input. Suspended solids (SS) content causes cloudiness of a fluid and can therefore be used to determine treatment efficiency through turbidity measurements. The main finding is that the ANN is able to learn how process input parameters influence the quality of the treated wastewater, and that combining machine learning with a technical understanding of the system provides valuable insight into operating wastewater treatment plants. Chemical dosage is currently determined by the inflow rate to the plant. However, it is known that flow pattern, reject flows and influent wastewater quality are important. The importance of these key factors were confirmed in the training process, which indicates that chemical dosage could be determined as a function of 1) SS concentration at the plant inlet, 2) the average hydraulic retention time (V/Q), 3) averaging the inflow parameters over a corresponding number of hours, and 4) whether or not the reject water pumps are on or off. Use of ANNs to design more advanced process control systems has the potential to increase chemical dosage precision and robustness of chemically enhanced primary treatment towards influent variations. This will be important for the degree of water purification and may lower the consumptions of chemical and thereby operational costs.