Self-organizing multi-modeling: A different way to design intelligent predictors

Identification of nonlinear systems is an important task for a large number of areas dealing with real-world applications and requirements. Recently, multiple works proposed “multi-model” based approaches to model nonlinear systems. Contrary to the conventional point of view, we propose to deem the multi-modeling as building a modular architecture, inspired from Artificial Neural Networks operation mode, where each neuron (module), represented by one of the local models, realizes some higher level transfer function. This article, deals with generalization of this new multi-modeling concept in the frame of nonlinear system's behavior identification and prediction context. Several multi-model construction strategies and identifiers issued architectures are presented and discussed. Experimental results validating presented multi-model based identifiers have been reported and discussed.