A Novel Protocol for Model Calibration in Biological Wastewater Treatment

Abstract

Activated sludge models (ASMs) have been widely used for process design, operation and optimization in wastewater treatment plants. However, it is still a challenge to achieve an efficient calibration for reliable application by using the conventional approaches. Hereby, we propose a novel calibration protocol, i.e. Numerical Optimal Approaching Procedure (NOAP), for the systematic calibration of ASMs. The NOAP consists of three key steps in an iterative scheme flow: i) global factors sensitivity analysis for factors fixing; ii) pseudo-global parameter correlation analysis for non-identifiable factors detection; and iii) formation of a parameter subset through an estimation by using genetic algorithm. The validity and applicability are confirmed using experimental data obtained from two independent wastewater treatment systems, including a sequencing batch reactor and a continuous stirred-tank reactor. The results indicate that the NOAP can effectively determine the optimal parameter subset and successfully perform model calibration and validation for these two different systems. The proposed NOAP is expected to use for automatic calibration of ASMs and be applied potentially to other ordinary differential equations models.

A consilience model to describe N2O production during biological N removal

“…Not all the model parameters are usually estimated from the available data as practical identifiability problems arise due to overparameterization of activated sludge models (ASM). 67 Model discrimination studies should therefore critically address calibration results as well as structural limitations. Best-fit parameter estimates provide little information and need to be supplemented with additional metrics (correlation matrix, sensitivity functions, analysis of residuals, estimation biases, etc.)…”

A Methodology for Global Sensitivity Analysis of Activated Sludge Models: Case Study with Activated Sludge Model No. 3 (ASM3)

“…A solution of the model may be written as y(t) = F(u, P, t), where u are external functions such as influent component concentrations, P is the vector of model parameters, t is time, F is the set of equations for ASM3, and y is the vector of model outputs. This study used an extended version of the base ASM3 evaluated by Zhu et al (2015) (see Supporting Information Appendix S1 for details of the model). The model consists of 14 dynamic biochemical components, 19 process rate formulations, and 43 kinetic and stoichiometric parameters.…”

“…3-ASM3/ASM3c) by the International Water Association (IWA; Henze, Gujer, Mino, & van Loosdrecht, 2000). Though activated sludge models such as the ASMs have been around for quite some time, an efficient model calibration and modification protocols have been a challenge (Sin, Van Hulle, De Pauw, van Griensven, & Vanrolleghem, 2005;Zhu et al, 2015). A key step in model calibration of activated sludge models is global sensitivity analysis (GSA) for the influence of model parameters on the model outputs (Cosenza, Mannina, Vanrolleghem, & Neumann, 2013).…”

“…Global sensitivity analysis measures the fluctuations of the time-dependent model outputs as affected by the perturbations in the model parameters. Parameters with high sensitivity indices ought to be prioritized for model calibration, and parameters with low sensitivity indices may be assumed invariant (Zhu et al, 2015). Several GSA approaches have been implemented on activated sludge models (Ji, 1993;Kim et al, 2006) such as Standard Regression Coefficient (SRC; Saltelli et al, 2008;Sin, Gernaey, Neumann, van Loosdrecht, & Gujer, 2011), Morris screening method (Cosenza et al, 2013), and Extended Fourier Amplitude Sensitivity Testing (Extended-FAST) method (Cosenza et al, 2013).…”

“…The implementation of these SA approaches, however, used the time-dependent variables (model outputs) as inputs for SA, which may be unsatisfactory due to the time-varying character of the GSA indices (Campbell, McKay, & Williams, 2006). Attempts to overcome the timevarying character of GSA indices calculated from time-series data include grand averages of the time-series indices (Zhu et al, 2015). Recently, however, mathematical works showed that time-dependent variables may be transformed using some basis functions to obtain time-independent representations of the model outputs to improve SA results (Campbell et al, 2006;Sumner, Shephard, & Bogle, 2012).…”