ANN_ddielectric - Conference Record of the 2002 IEEE...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
Conference Record of the 2002 IEEE lntemational Symposium on Electrical Insulation, Boston, MA USA, April 7-10,2002 The Use of some Paradigms of Neural Networks in Prediction of Dielectric Properties for High Voltage Liquid solid and Gas Insulations L.Mokhnache . A. Boubakeur Dept. of Electrical Eng.,Faculty of Engineering, High Voltage Laboratory, Ecole Nationale Polytechnique University of Batna; Baina, Algeria [email protected] hotmail.com - Abstract: The aim of this paper is to reduce the ageing experiment time and predict thermal ageing stress for longest time intervals using some paradigms of artificial neural networks. We present also the prediction of the breakdown voltage in a point-barrier-plane air gap versus the gap length INTRODUCTION Currently, the electric utility is being restructured in many countries, and as a consequence cost savings become more important than ever. One way of reducing expenditures is through proper monitoring and maintenance of power equipment. A major cause of failure for large transformers is the thermal ageing of the insulation materials [l]. Also, the dielectric properties of the electrical insulation of high voltage (HV) cables are greatly affected by thermal ageing. Studies on thermal ageing of PVC used in cables produced in CABEL (a National cables company) and of transformer oil (BORAK22) used by SONELGAZ (National Company of Electricity and Gas) were carried out respectively under the supervision ENP HV Laboratory in CABEL and SONELGAZ laboratories. These experiments are costly and time consuming. The aim of this paper is to reduce the ageing experiment time and predict thermal ageing stress for longest time intervals using artificial neural networks (ANN). Using a learning time of less than 2000 hours for PVC and 1000 hours for BORAK22, the same non-linear characteristics obtained in the laboratory are predicted by the proposed ANN. Moreover, the ANN is able to predict the characteristics of BORAK22 at 3000 hours of ageing [2]. We present also the prediction of the breakdown voltage in a point-barrier-plane air gap versus the gap length for a switching impulse voltage forms using ANN. In this paper we have used three nets for prediction Prediction Methods We have developed three networks; two are trained by ROM and one by back-Propagation. ANNs are developed to predict thermal ageing for the PVC and BORAK22 and the breakdown voltage for a point-barrier-plane air gap. RBFG trained by Random Optimisation Method (ROM) had been applied in this paper, using two techniques: FFN pattern, and Batch learning technique. BP 182, El-Hkach, Algiers, Algeria [email protected] yahoo.com RBF trained by BP Several networks have been tested for prediction using BP, but there were too many parameters to manipulate including the number of hidden layers, learning rates and momentum, bias, etc. These networks suffer most of the time from local minima stagnation or trapping. The weights (w,) Gaussian centres (cjk) and variances (ojk) adjustments are one of the causes of local minima problem because of their dependence on the error gradient. Their adaptation is given by:
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

Page1 / 4

ANN_ddielectric - Conference Record of the 2002 IEEE...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online