A Machine Learning Based Tool for Voltage Dip Classification

Abstract

A Machine Learning based tool is presented in order to make voltage dips (VD) ex-post analysis more automatic and effortless. The tool takes as input the full waveforms associated to voltage dips occurring in the Italian MV networks and recorded by QuEEN monitoring system implemented by RSE. The first tool has been developed to classify events on the base of their HV/MV origin since the utilities will be responsible only for the events due to faults occurred in their networks; it uses the self-tuning Kalman Filter and Support Vector Machine (SVM) for extracting the VD’s features and classifying the events, respectively.Instead, the second tool, based on end-to-end Deep Learning techniques, has been developed to distinguish between “true” and “false” VD; it utilizes a Convolutional Neural Network (CNN) whose first layers undertake the task of the features extraction while the last layers carry out the events classification.