Contournement des isolateurs pollués

Abstract

This work aimed on the proposal a new formulation of the resistance of pollution layer allowing to improve the dynamic models, the estimate of the parameters of the arc by particle swarm optimization (PSO) and the prediction of the critical flashover voltage by neural fuzzy approach named adaptive neural-fuzzy inference systems (ANFIS) and the hybrid approach based on the least squares support vector machine (LS-SVM) and the heuristics PSO. A mathematical model of Topali based primarily on the geometrical characteristics of the insulator is adopted to test the performances of several intelligent approaches. Firstly, we propose the PSO approach to find the parameters of the arc. The validation of the approach was justified by comparing the results provided by this latter and those found by the genetic algorithms (GA) method. Secondly, we propose two predictive hybrid approaches, namely neural-fuzzy approach (ANFIS) of the type Takagi Sugeno (TK) and the hybrid approach based on LS-SVM with radial basis function (RBF) kernel and PSO for the prediction of the critical flashover voltage. Basing on the indices of validation RMSE, MAPE and R , satisfactory and more accurate results are obtained by using LSSVM-PSO to estimate the critical flashover voltage for the considered conditions compared with the artificial neural networks (ANN), fuzzy logic (FL) and ANFIS. A new formulation of the resistance of pollution layer is proposed to improve a direct current (DC) dynamic model. This latter take into account the open shape of the real insulator and the influence of the constriction of the current lines in the pollution layer. The improved dynamic model is validated by using experimental and theoretical results suggested in the literature.