Diagnostic du comportement diélectrique d’un transformateur de puissance à base de modèles mathématiques optimisés par la technique PSO Jury de soutenance : Nom et prénom Grade Qualité DJEKIDEL Rabah PROFESSEUR Président MAHI Djillali PROFESSEUR Encadreur BESSEDIK Sid Ahmed PROFESSEUR Examinateur SAYADI Ahmed M.C.B Co Encadreur . Promotion: 2021/

Abstract

Nowadays, Industry observes a change in its maintenance policy by seeking to reconcile the rationalization of costs with the reliability of installations. Technical assistance, such as the diagnosis of power transformers has become necessary. Diagnosing the condition of a transformer is based on two approaches: one basic, relying on a set of measurements relating to routine maintenance; the other expert, using methods that consist in evaluating the effects of failure modes affecting dielectrics. The objective of this work is predictive modeling by multiple regression of the degree of polymerization (DPv). This modeling uses three parameters: the aging time, the operating temperature and the concentration of the chemical compound 2-furfural. Regarding the modeling and prediction of the traction index (Tidx), simple regression has proven to be sufficient. To do this, we used data from measurements made on Kraft paper impregnated in Luminol oil, used as solid insulation in power transformers, under the influence of three levels of applied temperature: (150°C, 170°C and 190°C). The implementation of a particle swarm optimization algorithm or PSO (Particle Swarm Optimization) made it possible to estimate the coefficients of the model. The approach recommended and interpreted in terms of error shows a good match with the experimental readings. This justifies the effectiveness of this technique. We also believe that monitoring the aging of the paper by the regression model including the PSO optimization method makes it possible to obtain a higher correlation factor. It is efficient and more efficient than RNA.