Training of neurofuzzy power system stabilisers using genetic algorithms

The problem of selecting and tuning the parameters of a neurofuzzy controller using genetic algorithms is discussed in this paper. The neurofuzzy controller is implemented as a multilayer perceptron, in which the weights are fuzzy membership functions. The optimal values of the parameters of the if-part and the then-part membership functions have been found during the learning method by applying an appropriate fitness function based on the controlled plant output. The proposed method has been applied to optimise the parameters of a neurofuzzy power system stabiliser (NF PSS). The overall system has been tested on a simulation model in different operating conditions and improved responses have been achieved.