Hybrid fuzzy modelling using simulated annealing and application to materials property prediction

Abstract

Proposes a hybrid fuzzy modelling approach using a self-organising network and simulated annealing algorithm for self-constructing and optimising fuzzy rule-based models. The proposed fuzzy modelling procedure consists of two stages. Firstly, a fuzzy competitive neural network is exploited as a data pre-processor to extract a number of clusters which can be viewed as an initial fuzzy model from engineering data. This step is used to perform fuzzy classification with the objective of obtaining a self-generating fuzzy rule base. Secondly, simulated annealing (SA), a combinatorial optimisation technique, is used to optimise the fuzzy membership functions. The application of this approach to the mechanical property prediction for C-Mn-Nb steels is given as an illustrative example.