Radial basis function based identifiers for adaptive PSSs in a multi-machine power system

Effectiveness of self-tuning adaptive power system stabilizers (APSSs) has been demonstrated in the literature. A self-tuning control algorithm consists of an identifier and a controller. As the algorithmic based identifiers require a relatively long computation time, investigations are being conducted to replace these with neural network based identifiers. In an interconnected power system, the generating units have different characteristics and sizes necessitating different pre-trained identifiers. Data gathered at various operating conditions and disturbances are used to train the RBF identifiers. The standard procedure for obtaining the centers for RBF identifiers involves initialization of the centers to random points in the input space. The centers are then updated using recursive rules. One limitation of this procedure is that even though the RBF centers learn the distribution of the input vectors by minimizing the distance between input vectors and the centers, they take long training time or fail to learn the topology of the input vectors. To overcome this limitation, an algorithm that creates RBF centers one at a time is described in this paper. Simulation studies on a five machine power system illustrate the effectiveness of this approach to the design of APSSs