Hybrid Adaptive Neural-Fuzzy Algorithms Based on Adaptive Resonant Theory with Adaptive Clustering Algorithms for Classification, Prediction, Tracking and Adaptive Control Applications

Akpan, Vincent A. and Agbogun, J. B. (2022) Hybrid Adaptive Neural-Fuzzy Algorithms Based on Adaptive Resonant Theory with Adaptive Clustering Algorithms for Classification, Prediction, Tracking and Adaptive Control Applications. American Journal of Intelligent Systems, 12 (1). pp. 9-33.

Text HYBRID ADAPTIVE NEURAL FUZZY ALGORITHMS BASED ON ADAPTIVE RESONANT THEORY Akpan and Abogun 2022.pdf Download (1MB)

Abstract

The development of a single compact algorithm for clustering, classification, prediction, tracking and adaptive control applications is currently of great challenge in the computational intelligence and adaptive control communities. This paper presents a new hybrid adaptive neural-fuzzy algorithm based on adaptive resonant theory with adaptive clustering algorithm (HANFA-ART with ACA) for clustering, classification, prediction, tracking and adaptive control applications. The HANFA-ART with ACA consist of nine major components, namely: (i) Mamdani fuzzy-type model; (ii) Takagi-Sugeno fuzzy-type model; (iii) ANN; (iv) FRBL; (v) adaptive resonant theory (ART); (vi) adaptive clustering algorithm (ACA) which is made up of (a) K-means clustering as the initialization algorithm, (b) adaptive mountain climbing clustering (AMCC) algorithm used in conjunction with Mamdani fuzzy-type model, and (c) adaptive Gustafson and Kessel clustering (AG-KC) algorithm used in conjunction with Takagi-Sugeno fuzzy-type model; (vii) Modified Leveberg-Marquardt algorithm (MLMA); (viii) adaptive recursive least squares (ARLS) algorithm; and (ix) several classes of membership function. The integration of these nine components results in the HANFA-ART with ACA which have been applied for the (i) classification of related diseases for the cause of meningitis using Mamdani fuzzy-type model and (ii) setpoint determination for activated-sludge waste water treatment plant (AS-WWTP) influent pump output predictions, tracking and adaptive control. Simulation results demonstrates the efficiency of the HANFA-ART with ACA when compared to standard adaptive neuro-fuzzy inference system trained with back-propagation with momentum (ANFIS with BPM) and proportional-integral-derivative (PID) control algorithms based on some evaluation criteria. The HANFA-ART with ACA can be adapted and deployed for extensive clustering, classification, prediction, tracking and adaptive control applications without explicit process model as justified in this work.

Item Type:	Article
Subjects:	Q Science > Q Science (General)
Divisions:	Faculty of Natural and Applied Sciences
Depositing User:	mrs chioma hannah
Date Deposited:	22 Nov 2022 11:20
Last Modified:	22 Nov 2022 11:20
URI:	http://eprints.gouni.edu.ng/id/eprint/3921

Actions (login required)

View Item