Recursive Identification and Adaptive Forecasting of Wastewater Flows

Abstract

In the control of wastewater pumping stations and treatment plants. the forecasting of sewage flows will allow better management of the sewer system by reducing overflows and operational cost. Wastewater flows normally exhibit a diurnal flow pattern with large variations in daily mean and are affected by rapid rainfall infiltration and the slower groundwater infiltration. These factors have to be taken into consideration in the prediction method.A direct k-step adaptive predictor is used to forecast the wastewater flow. The parameters of the multi-input/single output ARMAX model are recursively estimated at each time step by the method of extended least squares; a forecast of the wastewater flow k-step ahead is then made on the basis of the updated model. This method is illustrated for a sewage catchment in Melbourne. Australia which has a sewer system separate from the stormwater system.The model uses the measured sewer flow as outputs and rainfall intensities and/or flow pattern as inputs. Prefiltering and averaging of the data prior to identification reduces the error in the estimation of the fast response peaks in wet conditions. Under normal flow conditions where the rate of increase in sewer flow is slow. the adaptive predictor is robust. In wet conditions. the estimation of the size and occurrence of the fast response peak was found to be dependent on the structure of the model.