APLICACIÓN DE ALGORITMOS DE APRENDIZAJE AUTOMÁTICO EN LA PREDICCIÓN DE FALLAS EN REDES ELÉCTRICAS

Abstract

The early prediction of failures in electrical networks using machine learning techniques constitutes a field of research of growing interest given its ability to improve the reliability and efficiency of electrical systems. The present study evaluates and compares the performance of supervised and unsupervised algorithms, including artificial neural networks (ANN), random forest (RF), support vector machines (SVM) and principal component analysis (PCA). The results indicate that ANN and RF achieve the best results with an accuracy of 92% and 89% respectively. ANNs excel in sensitivity (90%) while RF and SVM achieve greater specificity (90%). The area under the ROC curve confirms ANN as the most effective classifier (0.95), followed by RF and SVM (0.94 both). The analyzes also reveal a direct relationship between data volume and model performance. Likewise, data quality significantly impacts accuracy. In conclusion, deep learning techniques, specifically ANN, constitute promising alternatives for the implementation of intelligent electrical grid management systems. However, data curation and the combination of complementary approaches are still necessary to strengthen the models.