Abstract:

this paper presents a novel approach to mitigate MATLAB’s ANFIS constraint, which allows only for the use of Sugeno method without rule sharing when developing a neuro-fuzzy system to evaluate complex systems. The proposed approach relies on building a hierarchical multi-level Sugeno FIS models. The hierarchical model eliminates the need for a more complex Mamdani model or the need to use a Mamdani FIS short of being able to train them due to the rule sharing constraint in ANFIS.  The proposed approach has proven to be more effective than simple Sugeno ANFIS models, with very low error rates. We use the “tipping problem” as a case study with two measurable input variables and three outputs to illustrate method operability and its effectiveness.