A new approach for generating a system model from its input-outputdata is presented. The model is approx- imated as a linearcombination of simple basis functions. The number of basis functionsis kept as small as possible to pre- vent over-fitting and to makethe model efficiently computable. Based on these conditions, geneticprogramming is employed for the generation and selection o theappropriate basis. Since the obtained model can be expressed insimple mathematical expres- sions, it is suitable for using the modeas a macro or behav- ior model in system level simulation.Experimental results are shown.
展开▼
