A neural power system stabilizer trained using local linear controllers in a gain-scheduling scheme

This paper describes the design of a Power System Stabilizer synthesized by a static Artificial Neural Network. The patterns used to train the neural network are sets of controller parameters, previously calculated for several system operation points using a pole-shifting method. This neural network stabilizer then operates, in a gain-scheduling scheme, in accordance with the values of active and reactive powers furnished by the generator to the power system, but with soft transition variations of the controller parameters. The trained neural network presents, as its main characteristic, almost uniform values for all the stabilizer parameters when the system synchronous machine is generating reactive power, but these same parameters suffer great variations when the machine is absorbing reactive power. Simulation tests presented show very good performance for the proposed Neural PSS, when compared with a fixed-parameter stabilizer, corroborating the main characteristic of the proposed stabilizer.