Rahim Khorsandi Gavgani

  Abstract

  Electric power transmission network are near different source of pollution so the insulators surface leakage current increases and probability of flashover can increased. During the past years several criteria to assess contamination has been introduced, one of the most common is ESDD . There are Different methods to determine ESDD . Some of them include regression - based method to find mathematical relationships between different parameters which can important effect on pollution . These methods some times haven’t any answer, because during the mathematical operations to calculate the amount of ESDD, in some cases it is necessary to obtain the inverse matrix but matrix’s Determinant is zero. The Other methods for determination of ESDD use sampling measurement of actual contamination there for they are time consuming and expensive .

  In order to solve the mentioned problems, using a smart method is appropriate because these methods typically have suitable speed and accuracy . Artificial neural network has been selected so we don’t need obtaining of mathematical relationship between the parameters. In this thesis, using a MLP neural network structure and change the number of hidden layer neurons ESDD levels with appropriate accuracy is obtained. Simulations results show that the number of layers of neural network have important influence on accuracy of ESDD prediction. Amount of pollution divided to four group low, medium, high and very high. Artificial neural network can online predict the ESDD of polluted insulator. During prediction of ESDD by ANN, some parameters such as temperature, wind velocity, insulator dimension and rain have more influence. Normalization of data between -0.9 to 0.9 causes that effect of all data calculated in prediction so accuracy of obtained ESDD value is better than references.

  Defence Date:7 july 2010 Time:10:12

  Location:Electrical Engineering Department