Гібридні алгоритми самоорганізації моделей для прогнозування складних процесів

Abstract

A new kind of GMDH algorithm of hybrid type is proposed combining both multilayered and combinatorial schemes for model self-organization and includes the following main novelties: using initial arguments in each layer to avoid loosing the relevant ones; optimization of every partial model structure by combinatorial algorithm to avoid overfitting.