A neural network model to predict the wastewater inflow incorporating rainfall events

Under steady-state conditions, a wastewater treatment plant usually has a satisfactory performance because these conditions are similar to design conditions. However, load variations constitute a large portion of the operating life of a treatment facility and most of the observed problems in complying with permit requirements occur during these load transients. During storm events upsets to the different physical and biological processes may take place in a wastewater treatment plant, and therefore, the ability to predict the hydraulic load to a treatment facility during such events is very beneficial for the optimization of the treatment process. Most of the hydrologic and hydraulic models describing sewage collection systems are deterministic. Such models require detailed knowledge of the system and usually rely on a large number of parameters, some of which are uncertain or difficult to determine. Presented in this paper, an artificial neural network (ANN) model that is used to make short-term predictions of wastewater inflow rate that enters the Gold Bar Wastewater Treatment Plant (GBWWTP), the largest plant in the Edmonton area (Alberta, Canada). The neural model uses rainfall data, observed in the collection system discharging to the plant, as inputs. The building process of the model was conducted in a systematic way that allowed the identification of a parsimonious model that is able to learn (and not memorize) from past data and generalize very well to unseen data that was used to validate the model. The neural network model gave excellent results. The potential of using the model as part of a real-time process control system is also discussed.