ANN Ensemble and Output Encoding Scheme for Improved Transformer Tap-changer Operation

Voltage control of power transmission and distribution system using artificial neural network (ANN) based tap-changer control has a number of potential advantages when the parallel transformers are connected across the power network. Previous development of ANN based tap-changer control were made using modified cascade correlation learning algorithm incorporating the Bayesian framework and produced above 99% correct tap-changer operation in average. This paper investigates and exploits a suitable output coding and ensemble principle in the design of ANN based tap-changer control to further enhance its performance producing more than 99.95% correct tap-changer operation. Ultimately, the ensemble design makes the ANN based control more reliable. Performances of ANN ensembles for tap-changer control are analyzed and results are presented