The Ordinary Kriging in Multivariate Dynamic Modelling and Multistep-Ahead Prediction

This paper investigates the data-driven MultiVariate Dynamic Modelling (MVDM) and MultiStep-Ahead Prediction (MSAP) of nonlinear systems based on the Ordinary Kriging (OK) metamodel. The OK is used to build a set of Nonlinear Autoregressive models with Exogenous inputs (NAREX), able to approximate the system future outputs as a function of previous inputs and outputs considering a specific delay. Then, these OK-based dynamic models are used in a recursive interactive interpolations scheme to predict the process outputs over several time steps. The capabilities of the OK-based dynamic models are compared to other leading techniques, via their application to benchmark cases. The application results reveal the OK promising and competitive capabilities for MVDM of nonlinear systems, in terms of prediction accuracy and prediction time horizon.