What Else Do the Deep Learning Techniques Tell Us about Voltage Dips Validity? Regional-Level Assessments with the New QuEEN System Based on Real Network Configurations

Abstract

The paper presents the performance evaluation of the DELFI (Deep Learning for False voltage dip Identification) classifier for evaluating voltage dip validity, now available in the QuEEN monitoring system. In addition to the usual event characteristics, QuEEN now automatically classifies events in terms of validity based on criteria that make use of either a signal processing technique (current criterion) or an artificial intelligence algorithm (new criterion called DELFI). Some preliminary results obtained from the new criterion had suggested its full integration into the monitoring system. This paper deals with the comparison of the effectiveness of the DELFI criterion compared to the current one in evaluating the events validity, starting from a large set of events. To prove the enhancement achieved with the DELFI classifier, an in-depth analysis has been carried out by cross-comparing the results both with the neutral system configuration and with the events characteristics (duration/residual voltage). The results clearly show a better match of DELFI classifications with network and events characteristics. Moreover, the DELFI classifier has allowed us to highlight specific situations concerning power quality at regional level, resolving the uncertainties due to the current validity criterion. In details, three groups of regions can be highlighted with respect to the frequency of the occurrence of false events.