A deep learned type-2 fuzzy neural network: Singular value decomposition approach

Abstract

The main objective of this study is to present a novel dynamic fractional-order deep learned type-2 fuzzy logic system (FDT2-FLS) with improved estimation capability. The proposed FDT2-FLS is constructed based on the criteria of singular value decomposition and uncertainty bounds type-reduction. The upper and the lower singular values of the set of inputs are estimated by a simple filter and the output is obtained by fractional-order integral of the uncertainty bounds type-reduction. Using stability criteria of fractional-order systems, the adaptation rules of the consequent parameters are extracted such that the globally Mittag-Leffler stability is achieved. The proposed FDT2-FLS is employed for online dynamic identification of a hyperchaotic system, online prediction of chaotic time series and online prediction of glucose level in type-1 diabetes patients and its performance is compared with other well-known methods. It is shown that the proposed mechanism results in significantly better prediction and estimation performance with less tunable parameters in just one learning epoch.