A PPLICATION SUR L’ALGORITHME ISOLATION FOREST POUR LA DÉTECTION D’INTRUSIONS DANS UN RÉSEAU ÉLECTRIQUE BASSE TENSION DE KASANGULU

Abstract

The reliability of the low-voltage electricity network in the DRC, and particularly the electricity network in Kasangulu, where infrastructure is vulnerable to excessive consumption, fraud and tampering with electricity meters. Traditional detection methods, based on fixed thresholds, prove ill-suited to the variability of consumption profiles and the rarity of anomalies, which makes detection more complex. In this context, this article proposes an intrusion detection approach based on the unsupervised Isolation Forest algorithm, applied to a subscriber’s load data, specifically the energy consumed, time slots and the duration of exceedances, in order to predict and detect anomalies in the Kasangulu LV power grid. The approach was developed and its performance evaluated using the confusion matrix, the ROC curve and temporal analysis. This highlights the algorithm’s ability to effectively detect anomalies without using labelled data. The results demonstrate that iForest is a suitable solution for the proactive monitoring of the LV power grid; with an average intrusion detection capability (AUC of 60%), it helps to enhance the security and reliability of the Kasangulu LV power grid.