Fuzzy rules extraction based on output-interval clustering and support vector regression for forecasting

Abstract

This paper presents a fuzzy rules extraction algorithm based on output-interval clustering and support vector regression. The approach is unlike most existing clustering algorithms for structure identification of fuzzy systems, where the focus is on combined input–output clustering. The output-interval clustering algorithm divides the output space into several partitions and each output partition is considered to be an interval; then, input data are projected into sub-clusters that are based on the input distribution constrained by the output intervals. Fuzzy rules are extracted from sub-clusters within each output interval. In order to have a more compact final system structure and better accuracy, local functions associated with each of the sub-clusters based on support vector regression are constructed. The fuzzy rule-based modeling scheme gradually adapts its structure and rules antecedent and consequent parameters from data. Its main purpose is continuous learning, and adaptation to unknown environments. To illustrate the effectiveness of the approach, the paper considers a 2-D nonlinear function approximation, chaotic time series prediction and an operation learning application of steel mechanical property forecasting.