Determination of the response of Ar+SF6 to crossed electric and magnetic fields using an artificial neural network

AKCAYOL M. A., Hiziroglu H. R., Dincer M. S.

Annual Conference on Electrical Insulation and Dielectric Phenomena, Vancouver, Canada, 14 - 17 October 2007, pp.695-697 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • Doi Number: 10.1109/ceidp.2007.4451632
  • City: Vancouver
  • Country: Canada
  • Page Numbers: pp.695-697
  • Gazi University Affiliated: Yes

Abstract

In this study, an artificial neural network (ANN) is proposed to predict the mean energy and deflection angle that cause a breakdown in Ar+SF6 mixtures under crossed electric and magnetic fields. The selected ANN structure for this study is a fully connected hierarchical network consisting of an input layer, a hidden layer and an output layer. To train the ANN, results from a Monte-Carlo simulation have been used. The activation function for neurons is a sigmoid function with 0.5 threshold value. The predictions have R-2-values equal to 0.998 for epsilon and 0.9998 for theta. The relative error between the results of the Monte Carlo simulation and the predicted values of mean energy and deflection angle using the ANN is found to be less than 10%.