Keywords

glue, parameter estimation, bioprocess model

Start Date

1-7-2012 12:00 AM

Abstract

Uncertainty analysis in integrated wastewater treatment modelling is still in its infancy, although techniques from different fields are increasingly used in research for design and control of wastewater treatment plants (WWTPs). However, results should be interpreted with care. This paper shows explicitly the influence of different methodologies and subjective choices on prediction uncertainty for a simple environmental modelling case: a respirometric experiment for acetate degradation without storage. This case uses experimental data and can be used to estimate kinetic parameters of activated sludge in full-scale wastewater treatment modelling. These are subsequently used in more complex activated sludge models (ASMs) for dynamic modelling studies, in order to reduce the number of parameters to calibrate in these overparameterised models. Three uncertainty analysis methodologies are compared: (1) classical parameter estimation (Fisher Information Matrix or FIM-based), which has been performed extensively for respirometric models. The derived parameter confidence intervals are then propagated to the output through Monte Carlo (MC) simulations with and without correlation based sampling; (2) MC simulations from expert-based probability density functions; (3) the generalised likelihood uncertainty estimation (GLUE) method, a well-known uncertainty analysis method applied in hydrological modelling, recently also applied on full-scale WWTPs. The output uncertainty boundaries on the model output are very specific to the method used and to subjective choices like probability density functions in the expert-based method and threshold values in the GLUE method. Besides, classical parameter estimation and GLUE are useful to observe and easily handle correlated parameters, which is very important when using these methods on full-scale models. Appropriate validation experiments are needed to judge the applicability of the different applied methods. Whereas this seems not possible at full-scale level, it should be feasible to perform validation respirometric experiments at lab-scale.

COinS
 
Jul 1st, 12:00 AM

Influence of Uncertainty Analysis Methods and Subjective Choices on Prediction Uncertainty for a Respirometric Case

Uncertainty analysis in integrated wastewater treatment modelling is still in its infancy, although techniques from different fields are increasingly used in research for design and control of wastewater treatment plants (WWTPs). However, results should be interpreted with care. This paper shows explicitly the influence of different methodologies and subjective choices on prediction uncertainty for a simple environmental modelling case: a respirometric experiment for acetate degradation without storage. This case uses experimental data and can be used to estimate kinetic parameters of activated sludge in full-scale wastewater treatment modelling. These are subsequently used in more complex activated sludge models (ASMs) for dynamic modelling studies, in order to reduce the number of parameters to calibrate in these overparameterised models. Three uncertainty analysis methodologies are compared: (1) classical parameter estimation (Fisher Information Matrix or FIM-based), which has been performed extensively for respirometric models. The derived parameter confidence intervals are then propagated to the output through Monte Carlo (MC) simulations with and without correlation based sampling; (2) MC simulations from expert-based probability density functions; (3) the generalised likelihood uncertainty estimation (GLUE) method, a well-known uncertainty analysis method applied in hydrological modelling, recently also applied on full-scale WWTPs. The output uncertainty boundaries on the model output are very specific to the method used and to subjective choices like probability density functions in the expert-based method and threshold values in the GLUE method. Besides, classical parameter estimation and GLUE are useful to observe and easily handle correlated parameters, which is very important when using these methods on full-scale models. Appropriate validation experiments are needed to judge the applicability of the different applied methods. Whereas this seems not possible at full-scale level, it should be feasible to perform validation respirometric experiments at lab-scale.