Identification of aircraft dynamics using a SOM and local linear models

The self-organizing map (SOM) is a powerful tool to produce topology preserving subspace mappings of high dimensional data. In this work we combine a SOM with a set of local linear models to implement functional mappings and identify potentially nonlinear plants. The large dimensionality of the spaces involved (many degrees of freedom and large dynamic range of parameters) is an issue, hence we compare fixed versus growing SOM topologies. The performance of the proposed algorithms is tested on the simulated data obtained from a realistic aircraft model.