Coherent grouping of power systems for use in training artificial neural networks

This paper presents a methodology for applying artificial neural networks to power systems of various sizes while addressing the problem of increasing training set size with increasing power system size. A slow-coherency based network partitioning technique is used to group the generators and load buses of the 10-machine, 39-bus system into coherent areas. Next we use characteristic parameters of each area as input features to train and perform estimations using a feed-forward neural network.<
>