Time Series Methods and Alternative Surrogate Modelling Approaches

As discussed in Chap. 3 , one of the main applications of machine learning is in the development of surrogate models, which are computationally cheap compared to the original physics-based simulations. The most common way to develop a surrogate model is to use machine learning, but there are two main alternatives, namely multi-fidelity and reduced-order models. The first of these can rely heavily on machine learning, while the latter does not usually involve any machine learning although it does rely on data from the original model. Machine learning can be introduced when the problem is parameter dependent and/or nonlinear. Reduced-order models are considered intrusive in that modifications to the original model or numerical formulation are required.