Steady-state detection, data reconciliation and machine learning for hybrid process modelling

In the process industry, it is possible to encounter systems whose behavior cannot be mapped through a first principles (white-box) model. Hybrid models aim at integrating data- driven (black-box) elements within white-box process models in order to fill the gap between the white-model model predictions and the actual system response. The goal of this Thesis is to propose and implement a hybrid modelling framework, and to assess its performance with respect to a white-box model.