DescriptionNonlinear dynamical simulation models are widely employed to make forecasts of future conditions for complex physical systems such as meteorology systems, cosmology systems and energy systems. In practice, our simulation models are far from perfect, all models suffer from built-in imperfections. Even if the model were perfect, challenges remains to identify the model initial condition and parameter values. The goal of the project is to apply machine learning approaches, for example tree-based methods and deep neural networks, to aid non-linear dynamical system modelling including 1) replace physical model for short term predictions; 2) correct systematic model errors; 3) approximate model local gradient efficiently for operational approaches.
PrerequisitesStatistical ModellingResources
|
email: Hailiang Du