The report describes the consolidation/validation of the DELFI RMS criterion for the classification of voltage dips according to their validity, a classifier based on supervised Deep Learning techniques, now already integrated in the QuEEN monitoring system of medium voltage distribution networks. Validation was achieved by cross-referencing its outcomes and those of the second harmonic criterion, which has always been active in the QuEEN system, with the neutral configuration data of the monitored networks and the event characteristics of the test set used for validation. The comparison methodology adopted provided ‘supervisory’ criteria that were less subjective than the validity assessments made by industry experts.

DELFI RMS provided results that were more consistent with both the behavior of the power grid—expected for networks operated with compensated neutral—and the characteristics of the events. DELFI is also able to resolve the uncertainties inherent in the second harmonic criterion, managing to classify even the so-called Non-Defined events that this criterion fails to classify, which are, moreover, of not insignificant number in networks operated with isolated neutral (29.8%). In particular, 93.4% of the Non-Defined events for the second harmonic criterion in networks with compensated neutral are evaluated as valid, i.e., true, voltage dips by DELFI.

Thanks to the greater homogeneity of its performance, DELFI has also made it possible to highlight certain specific Power Quality situations at the regional level. The report also describes the first version of an application for correlating events with service protocols, related to startups and trips of protections installed in UNARETI’s primary substations. The application was created downstream of the integration of this distributor’s national monitoring data into the QuEEN system, and in the future it can be used to validate all the supervised criteria implemented in QuEEN for classifying events according to both their validity and their origin in high- or medium-voltage networks.