Viviane Rodrigues Botelho

A New Approach for Practical Identifiability Analysis Applied to Dynamic Phenomenological Models

Abstract Identifiability analysis is an essential tool for parameter estimation of dynamic models. The objective of this technique is to verify which parameters of this model should be identified. This paper proposes a new methodology, based in control system theory applied to the local sensitivity analysis of the model. First, the sensitivity matrix is scaled by applying an optimization problem that minimizes the condition number of the system. Then, analyzing the NSRGA matrix of the system, the most appropriated subset of parameters to be estimated is selected. To assess the proposed methodology, a case study is presented.

Parameter Estimation of Models with Limit Cycle based on the Reformulation of the Objective Function

Abstract Many processes show limit cycles, meaning that the system presents oscillatory behavior. The parameter estimation of such kind of systems is not a simple task, due to the non-convexity of the optimization problem. This paper proposes the inclusion of a driving term based on the damping factor in the classical objective function formulation, reducing the non-convexity of the problem. This driving term is reduced after each iteration until its complete elimination, as the system starts to have oscillatory behavior close to the limit cycle. Two case studies illustrate the strengths of the proposed approach: the Jobses’s dynamic model for ethanol fermentation with Zymomonas mobilis and the Di Meglio et al. (2009) model, which represents the slugging flow regime appearing in vertical risers. The results showed that the proposed approach was able to ensure the oscillatory behavior, forcing the dynamical behavior of the system to produce the limit cycle.