Título: Leakage current monitoring in high voltage insulator strings with neural network optimized with genetic algorithm
Autores: Maria Gabriely Lima da Silva, Jander Rodrigo De Santana Vieira, Byron Leite Dantas Bezerra & Sergio Campello Oliveira
Resumo: Partial discharges and flashovers in insulator strings, among others, are directly related to pollution presence on their surfaces and the ambient humidity. A partial discharge monitoring system has been developed and implemented in some locations in the Northeast Region of Brazil. This system has recorded approximately two years of information on humidity, temperature and peak leakage current activity during its operation. From the data collected by this network, it is possible to identify variations in the activity of leakage current peaks, showing more intense activity, which is related to the increased risk of flashover. This article presents an analysis of leakage current data using the prediction of current peaks with LSTM (Long Short-Term Memory) and MLP (Multilayer Perceptron) neural networks optimized with a genetic algorithm. The optimized models undergo a comparative analysis through hypothesis tests, along with experiments exploring their adaptation and generalization capabilities on different datasets. The results show that the LSTM model outperforms the MLP model in predicting leakage current peaks, achieving a significant reduction. The LSTM model demonstrated remarkable ability to anticipate activity peaks with a low error rate, enabling forecasts five days in advance, even under variable weather conditions. In interregional tests, it achieved an MSE of 0.041 and a MAE of 0.008 when predicting unobserved data. These results enable maintenance teams to act proactively, mitigating lightning risks and optimizing the scheduling of transmission line cleaning operations.
Palavras-chave: insulators; flashover; leakage current; predict; hybrid systems.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1191498
Artigo em PDF: CBIC_2025_paper1191498.pdf
Arquivo BibTeX:
CBIC_2025_1191498.bib