System identification; Artificial intelligence; Neural network-based models; Interval type-2 fuzzy logic; Modified interval type-2 neuro-fuzzy network (MIT2FNN)

System identification is a crucial sphere of engineering dedicated to finding econo- mical yet accurate models for fully understanding how systems behave. In effectuating this aim, these models predict future behavior while enabling simulations for optimization purposes inclusive of parameter adjustments where necessary for enhanced performance levels.

However, what makes identifying systems challenging is the selection process regar- ding model structure choice and the estimation method used when making predictions concerning non-linearities present in complex phenomena affecting multiple variables. Nonetheless, experts have devised viable options toward precise modelling solutions by employing sophisticated techniques such as artificial intelligence algorithms or polyno- mial multi-model frameworks.

he proposed thesis offers an approach that fuses interval type-2 fuzzy logic together with neural network training skills towards producing a generalized structure that enables both local model selection combined modeling which permits approximating or forecas- ting the behavior of any given system.

The results were obtained using three case studies: the chaotic Mackey-Glass time equation, a furnace system, and a multisection tank system. The results of the pro- posed network for the approximation and prediction of these systems were compared with techniques from the literature, and the modified type-2 neuro-fuzzy interval network (MIT2FNN) showed lower mean squared error (MSE) values than the other techniques.