NEURON INSPIRED LEARNING RULES FOR FUZZY RELATIONAL STRUCTURES

Fuzzy relational equations are a suitable framework for fuzzy modelling. However their constructive resolution methods suffer from known drawbacks. In order to determine approximate solutions, both conventional and adaptive gradient based learning methods are proposed, namely for extended versions of fuzzy relational structures. Simulation results show that, for both learning methods, good approximate solutions are found. However the adaptive version shows considerably quicker convergence rates. The problem of process fuzzy identification using the proposed framework is then outlined, as an illustrative application.