An EOCA-based interpretable fuzzy modeling approach

In this paper, a novel EOCA-based interpretable fuzzy modeling approach is proposed to obtain the trade-off between interpretability and accuracy of the Takagi-Sugeno-Kang (TSK) model, i.e., the moderate compactness or the well approximation and generalization ability. By means of the presented Enhanced Objective Cluster Analysis (EOCA) algorithm, the fuzzy partition in the input space is inherently reduced, which strengthens their robustness to the initial clustering condition. Following, the initial fuzzy partition is iteratively expanded and optimized in order to balance the accuracy and the compactness of the model. Finally, the consequent parameters are quickly estimated by the least square method. The simulation results of the electrical application example show the power of the presented method.