Recurrent Mechanism and Impulse Noise Filter for Establishing ANFIS

In many real applications, building and updating adaptive neuro-fuzzy inference system (ANFIS) based on noisy measuring data sources need to be performed such that the filtering impulse noise (IN) from the initial datasets (IDSs) and establishing the ANFIS via the filtered IDS are carried out simultaneously. Focused on this purpose, in this paper, a novel recurrent mechanism as well as a solution for filtering IN based on Lyapunov stability theory is proposed to establish an adaptive online IN filter (AOINF). Using the AOINF, kernel fuzzy-C-means clustering method, and the least mean squares method, a cluster data space deriving from the filtered IDS is created to which the ANFIS is then formed. The recurrent mechanism executes filtering IN to build ANFIS and using the ANFIS as an updated-filter to filter IN synchronously until either the ANFIS converges to the desired accuracy or a stop condition is satisfied. Surveys, including identifying dynamic response of a magnetorheological damper via measuring datasets, are performed to evaluate the proposed method.