An approach for construction and learning of interval type-2 TSK neuro-fuzzy systems

In this paper, we propose an approach for construction and learning of interval type-2 TSK neuro-fuzzy systems. In the structure identification phase, we develop a self-constructing rule generation method to group the data into fuzzy clusters and extract initial fuzzy rules for creating an interval type-2 TSK fuzzy system. Then, we construct an interval type-2 TSK fuzzy neural network in the parameter identification phase and propose a hybrid learning algorithm to refine the parameters of initial fuzzy rules for higher precision. The hybrid learning algorithm is composed of the particle swarm optimization and a recursive SVD-based least squares estimator. Finally, we have a set of refined fuzzy rules. Compared with other approaches, experimental results have shown our approach produces smaller root mean squared errors and converges more quickly.