Artificial neural networks and genetic algorithm modelling and identification of arc parameter in insulators flashover voltage and leakage current

In this paper, we present an optimisation method based on genetic algorithms and artificial neural networks (ANN) experimental data from artificially polluted insulators for the determination of the arc constants and dielectric properties in the surface. The study of flashover phenomenon in polluted insulators has not yet been described accurately through a mathematical model. The definitions of arc constants are very difficult, which is created in the dry bands when the voltage exceeds its critical value. In this work, a pollution flashover generalised model is used. The obtained results show that the mathematical model with optimised arc constants simulates accurately the experimental data and corroborate the inverse relationship between flashover voltage and pre-flashover leakage current. For this purpose, an ANN was constructed in MATLAB and has been trained with several MATLAB training functions, while tests regarding the number of neurons, the number of epochs and the value of learning rate have taken place, in order to find which net architecture and which value of the other parameters give the best result. To validate our method an experimental tests for different insulators show very good agreement with the measured values and the computed ones.