Parameter estimation of actvated sludge model no. 1 from full scale plant input/output data

The Activated Sludge Model No. 1 developed by the International Association on Water Quality is well accepted for dynamic simulation of removal of carbonaceous and nitrous compounds from wastewater in activated sludge plants. In several important applications of the model, it is essential to obtain meaningful parameters to describe full scale plant behaviour accurately, which is referred to as model calibration. Several procedures for model calibration have been proposed and applied in the literature, ranging from obtaining parameters directly from observed input/output behaviour of plants to using — sometimes many-additional, dedicated experiments. We investigated the possibilities and limitations of using only typical input/output data of plants by establishing how many and which parameters can be estimated from these data, in order to determine which additional experiments are really needed. For this aim, important and identifiable parameter subsets were selected by sensitivity and identifiability analyses. Two different identifiability criteria were used, the determinant of the Fisher Information matrix and a combined criterion including both determinant and condition number. These criteria led to almost identical sets of parameters. The results of the analyses were tested by parameter estimation using simulated data without and with noise, as well as measured data from a full scale carrousel type treatment system. The simulated data allowed more parameters to be estimated accurately from the output set used than the measured data, which is most probably caused by model mismatch with the plant. With both simulated and real data, the parameter sets selected could be uniquely estimated from the data, indicating the usefulness of selecting identifiable parameter subsets.