Identifiability and physical interpretability of hybrid, gray-box models - a case study

Model identifiability concerns the uniqueness of uncertain model parameters to be estimated from available process data and is often thought as a prerequisite for physical interpretability model. Nevertheless, may challenging obtain in practice due both stochastic deterministic uncertainties, e.g. low variability, noisy measurements, erroneous structure, stochasticity locality optimization algorithm. For gray-box, hybrid models, rarely obtainable high number parameters. We illustrate through an industrial case study - modeling production choke valve petroleum well that preserved even non-identifiable models with adequate parameter regularization estimation problem. To this end, real scenario, it beneficial model's predictive performance develop over mechanistic flexibility higher. Modeling six wells on asset Edvard Grieg using historical show 35\% reduction median prediction error across comparing On other hand, developed are influenced by data. The findings encourage research into online learning variants improve results.