Forecasting the trajectories of dynamic systems for a long period of time using machine learning method

In the last decades, machine learning have been widely used to analyze the behavior of trajectories of dynamic system over long time intervals. Predicting the behavior of real systems does not allow the direct use of asymptotic methods due to the presence of many external forces and impossibility to accurately measure both the acting forces and the parameters of the system itself. Nevertheless, the methods of Uncertainty Quantification, Surrogate modeling, Blind Source Separation and others developed recently, together with various machine learning methods, often succeed in predicting trajectories for a long time interval.
The talk will describe applications of various dimensionality reduction methods to the prediction of satellite trajectories of global navigation systems.