Calculation of breakdown voltages in Ar+SF6 using an artificial neural network

Tezcan S., Dincer M., Hiziroglu H.

Annual Conference on Electrical Insulation and Dielectric Phenomena (CEIDP), Tennessee, Amerika Birleşik Devletleri, 16 - 19 Ekim 2005, ss.59-62 identifier identifier

  • Cilt numarası:
  • Doi Numarası: 10.1109/ceidp.2005.1560620
  • Basıldığı Şehir: Tennessee
  • Basıldığı Ülke: Amerika Birleşik Devletleri
  • Sayfa Sayıları: ss.59-62


An artificial neural network is proposed to predict the breakdown voltages in Ar+SF6 gas mixtures. The proposed neural network is designed with one hidden layer that includes twenty-five neurons. The output layer of the ANN consists of one neuron, which is essentially the predicted breakdown voltage. In order to train the ANN, the experimental data available for Ar+SF6 have been used. The results of this ANN are compared with the experimental data as well as calculated data using the streamer criterion. With the proposed ANN, the average relative errors on breakdown voltages are found to be 3.85% for training and 4.32% for testing. Since the average errors are less than 5%, it is recommended to use ANN to predict the breakdown voltages.