Applied unsupervised learning in model reduction of linear dynamic systems

In this paper, a method of unsupervised learning is proposed for the purposes of reducing large-scale complex dynamic systems. Reduction of a system is carried out through the division of state variables into groups and through the selection of the characteristic representatives of each group. The proposed methodology is tested on an electric power system. The obtained results indicate that the model of the dynamic system can be significantly simplified while retaining its basic dynamic characteristics.