Echo state network applied for classification of medium voltage insulators

Abstract

Insulators are components of electrical power grid that have the function of mechanically supporting cables and isolating electrical potential. The proper functioning of the insulators is essential for the continuity in the supply of electrical energy. When an insulator has its properties damaged, disruptive discharges may shut down the system and impairs the network’s reliability. For this reason, classifying adverse conditions is a critical task to keep the system running. In this paper, the echo state network is used for the classification of the insulators based on the ultrasound signal. Hypertuning is applied to automate the evaluation of the parameters and optimize the network to have a general application. The insulators are evaluated in the laboratory under controlled conditions from the application of 7.95 kV (phase-to-ground), under the same conditions that are found in the field. The assessment is made on perforated and contaminated insulators, which were removed from service due to defects. The echo state network achieves 87.36 % accuracy for the multiclassification and 99.99 % for the specific classification of drilling. For comparative analysis, the multilayer perceptron and support-vector machines are evaluated based on the fast fourier transform. The results show that echo state network is promising to classify the evaluated conditions, being more accurate than the multilayer perceptron and support-vector machines based on fast fourier transform.