Improving Wang-Mendel method performance in fuzzy rules generation using the fuzzy C-means clustering algorithm

The generation of fuzzy rules from samples for fuzzy modeling and control is significant If samples contain noise and outliers, the Wang-Mendel (WM) method may lead to the extraction of invalid rules resulting in low confidence of the rules. The scale of the samples also affects the efficiency of the WM method. Interaction among input variables can help the WM method achieve high completeness and robustness. The fuzzy C-means clustering (FCM) algorithm can reduce the scale of samples and undo noisy data to some degree. This paper aims to develop an FCM-based improved WM method that adopts a modified FCM algorithm to preprocess the original samples and compute the interaction among the samples. Then, the optimized samples are used to generate fuzzy rules, thereby building a complete rule set through extrapolation. Experimental results from two nonlinear functions and short-term load forecasting case study show that the proposed method not only has high completeness and robustness, but also ensures better prediction accuracy of the fuzzy system. (C) 2014 Elsevier B.V. All rights reserved.