Evaluation of Sensitivity and Observability of the State Vector for System Identification and Experimental Design

Abstract

This paper presents two new algorithms to evaluate numerically the state and coefficient's observability and sensitivity in dynamic models. The observability algorithm computes the numerical condition of observability matrix as an index of the ability to identify the unknown states or coefficients. The sensitivity algorithm quantifies dynamically the influence of unknown variables on measurable variables. Both algorithms are useful tools to predict the identification accuracy and to design the best operational conditions to estimate each one of the unknown parameters. The algorithms have been applied to discuss the identification of kinetic coefficients in the IAWPRC biodegradation model, using an experimental procedure based on changing the values of reactor volumes.