Neural networks for fast voltage prediction in power systems

In power system security assessment, the prediction methods allow fast local approximation of the numeric load flow algorithm. A preliminary contingency screening is then obtained by quick estimation of the postfault bus voltages and line power flows. In last decades artificial neural networks (ANN) proved to be very suited for approximation of complex input-output relationships, learnt from a set of samples. However, most proposed methods employ the classic multi-layer perceptron trained with the back-propagation algorithm, which lacks fast learning capabilities. In this paper, we present some different approaches to the fast voltage prediction task. For this purpose, we compare the capabilities of three fast learning ANNs (the radial basis function network, the progressive learning network, and the self-organising map used as associative memory). Simulation tests, referred to normal and post-fault conditions, have been carried out in a wide range of operating scenarios