Multilevel composite fuzzy models

The concept of multilevel fuzzy modeling is presented. It is realized as a sequence of k+1 identification procedures of one main fuzzy model and k correction fuzzy models. The resulting multilevel composite fuzzy model CFM is an additive structure of all k+1 models that has the ability to gradually improve the approximation accuracy by adding new correction models. The overall accuracy of this modeling approach highly depends on the identification accuracy of each particular sub-model. The multilevel fuzzy modeling could be used as one possible approach to decreasing the total number of parameters of the fuzzy model by its decomposition into a sequence of simpler fuzzy models. The concept of CFM can be also used with different data sets taken at different times. In such cases the CSM is able to update the overall model behavior according to the new process information still keeping the behavior learned by the previous data set. Finally the multilevel structure of the proposed model could be utilized even with different types of models, not necessarily fuzzy models only. This could be the case when the basic level model is a kind of analytical or stochastic model and the other (correction) level models are fuzzy models learned from the next available data sets. A simple test example in the paper demonstrates the technology of creating as well as the main features of the proposed multilevel fuzzy modeling