Reduced-Support-Vector-Based Fuzzy-Neural Model with Application to the Material Property Prediction

A fuzzy model based on support vector regression (SVR) and particle swarm optimization (PSO) for the property prediction of heat treatment process of alloy steels is presented in this paper. First, a SVR model is built and the parameters of SVR are optimized by using the grid optimization algorithm, a set of equivalent fuzzy IF-THEN rules is generated from the obtained support vectors, then PSO is utilized to obtain a optimal fuzzy model with reduced rule (support vector) which approximate pre images of the original SVR model. The proposed modeling approach has been used for the mechanical property prediction in hot-rolled steels. Preliminary results reveal that the proposed modelling approach can lead to accurate and flexible fuzzy models.