Hybrid Model of Interval Type-2 Neural Fuzzy Inference System and Mutual Subsethood with Applications

Abstract

This paper presents the hybrid network of Interval Type-2 Subsethood Neural Fuzzy Inference System (IT2SuNFIS) [1] and Interval Type-2 Mutual Subsethood Neural Fuzzy Inference System (IT2MSFuNIS) [2] and it’s applications in the area of Mackey-Glass time-series (MGTS) prediction, chemical plant control and Hang function approximation. This model introduces the mutual subsethood measure [2] in IT2SuNFIS network [1]. Resulting hybrid network differs from IT2MSFuNIS in its consequent network structure and it is different from IT2SuNFIS as well; it uses mutual subsethood measure instead of subsethood measure for determining correlation between the interval type-2 fuzzy set (IT2 FS) inputs and IT2 FS antecedents. The inputs to the system are fuzzified using IT2 FSs with Gaussian primary membership function (GPMF) having identical mean but different variance. The signal aggregation of type-2 based activation is performed using product operator. This hybrid neuro-fuzzy system learns using different differential evolution (DE) strategies and Artificial Bee colony differential evolution (ABC-DE) framework [3]. Empirical studies are conducted on benchmark data sets of MGTS, a chemical plant and Hang function. Comparisons with other type-1 and type-2 neuro-fuzzy models and in particular with IT2SuNFIS [1], verify the excellent performance of the hybrid network.