Gaussian process regression for modeling of time-varying systems

In recent times, data based methods have gained importance especially for complex and high dimensional systems due to a short development time compared to physical modeling. The disadvantage is however, that even simple knowledge such as extrapolation behaviour can in general not be considered during model generation. In this article, a modeling scheme based on Gaussian process regression is presented, which is able to incorporate such knowledge in the model. Moreover, in many systems an adaptation of the model to the individual system is necessary. To this end, an online adaptation scheme is presented, which uses the uncertainty information of the Gaussian process to detect changes in the model and incorporate them.