Self-organising fuzzy perceptrons applied to power system stability

Organising and adjusting a neuro-fuzzy system is been presented in this paper. A fuzzy inference system has been implemented on a multilayer perceptron, in which the weights are fuzzy membership. The parameters of the fuzzy multilayer perceptron are meaningful and have physical interpretation. A hierarchical procedure is proposed for design and organising the system in three levels: predefining the rules, adjusting the membership functions using a supervised learning and improving the behaviour of the system by unsupervised learning. The error back-propagation (EBP) method is used for adjusting the fuzzy weights. This system has been used for damping the electromechanical mode of oscillations, as a power system stabiliser (PSS). The rotor speed deviation and acceleration are used as the PSS inputs, which are converted to an angle and a magnitude in the phase plane. Some conditions have been proposed to facilitate the employment of the gradient decent method for adjusting the parameters of the fuzzy perceptron. The effectiveness of the proposed neuro-fuzzy PSS at different operating points of the power system and a comparison with other PSS are investigated by simulation studies.