Neural network assessment of small signal stability

Neural network (NN) techniques have been recently applied to several power system problems. So far the most common types of neural networks used in the power field have been the layered perceptron, trained with a backpropagation algorithm (supervised learning), and Kohonen´s self-organizing feature maps (unsupervised learning). This paper presents results obtained through the use of the first of these NN techniques, in the assessment of small signal stability of a single-machine infinite-bus power system. Certain variables of interest (real and reactive power) are considered as the inputs, while different types of outputs are chosen and the performance of the NNs compared for each of the cases. A layered perceptron is trained and employed as a classifier (stable or unstable point), and as a regression machine (interpolating a numerical value), depending on the kind of the output, in order to increase the speed of assessment of the stability of the system. A power system stabilizer (PSS) applied to the generator excitation system is also considered, and results are compared with the previous analysis