Reduction of fuzzy control rules by means of premise learning – method and case study

Rule number reduction is important for fuzzy control of complex processes with high dimensionality. It is stated in the paper that this issue can be treated effectively by means of learning premises with general structure. Since conditions of rules are generalised by a genetic algorithm (GA) rather than enumerated according to every AND-connection of input fuzzy sets, a parsimonious knowledge base with a reduced number of rules can be expected. On the other hand, to give a numerical evaluation of possible conflicts among rules, a consistency index of the rule set is established. This index is integrated into the fitness function of the GA to search for a set of optimal rule premises yielding not only good control performance but also little or no inconsistency in the fuzzy knowledge base. The advantage of the proposed method is demonstrated by the case study of development of a compact fuzzy controller to balance an inverted pendulum in the laboratory. [Copyright &y& Elsevier]