Improved calibration of a solid substrate fermentation model

Abstract

Background: Calibration of dynamic models in biotechnology is challenging. Kinetic models are usually complex and differential equations are highly coupled involving a large number of parameters. In addition, available measurements are scarce and infrequent, and some key variables are often nonmeasurable. Therefore, effective optimization and statistical analysis methods are crucial to achieve meaningful results. In this research, we apply a metaheuristic scatter search algorithm to calibrate a solid substrate cultivation model. Results: Even though scatter search has shown to be effective for calibrating difficult nonlinear models, we show here that a posteriori analysis can significantly improve the accuracy and reliability of the estimation. Conclusions: Sensibility and correlation analysis helped us detect reliability problems and provided suggestions to improve the design of future experiments.

Characterization of a Regenerative Hydrogen-Vanadium Fuel Cell Using an Experimentally Validated Unit Cell Model

“…A parametric sensitivity analysis was performed considering only well fitted points of the different charge-discharge curves (data set 3 to 9 and 15 to 21). 37,38 The average absolute sensitivity of the cell potential (|G ave i |) with respect to the fitting parameters (θ) was assessed as a function of current density. These average absolute sensitivities were obtained considering the sensitivity coefficient (G i ) as a function of time, 38 Equation 19.…”

“…37,38 The average absolute sensitivity of the cell potential (|G ave i |) with respect to the fitting parameters (θ) was assessed as a function of current density. These average absolute sensitivities were obtained considering the sensitivity coefficient (G i ) as a function of time, 38 Equation 19. Also, the normalized sensitivity coefficients (G nor i ) were calculated using a central finite difference approximation after running simulations for perturbation in each fitting parameter, 37,39 Equation 20.…”

Comparison of noninvasive, in-situ and external monitoring of microbial growth in fed-batch cultivations in Corynebacterium glutamicum

“…The variables were normalized taking into account that the sensor signals and the optical density have different orders of magnitude. Post-regression diagnostics are techniques that test the quality of the parameter estimation [34]. The methods determine the significance of the confidence intervals, the identifiability with cross correlations between the parameters, and the sensitivity of the output response under parameter perturbations [35].…”