Lyu, W; Galvanin, F; (2024) DoE-integrated Sparse Identification of Nonlinear Dynamics for Automated Model Generation and Parameter Estimation in Kinetic Studies. Computer Aided Chemical Engineering , 53 pp. 169-174. 10.1016/B978-0-443-28824-1.50029-6.

Text Lyu et al (2).pdf - Accepted Version Access restricted to UCL open access staff Download (409kB)

Abstract

Digital twins have revolutionized manufacturing by utilising robust kinetic models to predict the behaviour of biochemical reaction systems under variable operating conditions. Identifying accurate expressions for reaction system models formulated as sets of differential and algebraic equations (DAEs) is challenging due to numerous state variables and kinetic parameters, compounded by limited observations and experimental errors. To enhance reliability, model structure confirmation should precede parameter estimation and validation. Traditional model-building approaches require prior knowledge of candidate models, whereas model generation methods like Sparse Identification of Nonlinear Dynamics (SINDy) only require the definition of a library of function terms. In this paper, we propose a new model identification framework, named Design of Experiment-integrated Sparse Identification of Nonlinear Dynamics (DoE-SINDy) to iteratively generate, evaluate, and select candidate models to represent systems with minimal training data. Tested on a simulated case study, DoE-SINDy succeeds in identifying an assumed true model efficiently with a limited experimental budget and noisy datasets.

Download activity - last month

Export as XMLJSONCSV

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item