Adaptive Neuro-Fuzzy Inference System Control for a Two Tanks Hydraulic System model

Abstract

This paper presents an Adaptive Neuro-Fuzzy Inference System (ANFIS) Control design for a Two Tanks Hydraulic System (TTHS) model. The ANFIS control algorithm was trained to perform as a successful Mamdani Fuzzy Logic Control (FLC) algorithm designed previously. Its performance was tested for different references as well as perturbation scenarios and compared with the FLC obtaining similar results. ANFIS and FLC design prove their superiority when compared to classical control algorithms because of their ability to deal with non-lineal systems. ANFIS is a modeling technique that provides an easy to implement and compute control algorithm provided that its training is adequately performed. This modeling technique takes advantage of the automatic tuning capabilities of Adaptive Neural Networks (ANN) and the ability to define linguistic variables that translate to human expertise like in FLC. FLC bases its functioning in the principle of heuristics so design is natural and easy to achieve. Only a basic knowledge of the system dynamics is needed for a successful design, and no mathematical model is needed for the design process. Results show an adequate performance of both control algorithms. The paper concludes by pointing out the benefits of translating the previously defined FLC algorithm into an ANFIS architecture.