Detection of electric vehicle charging events in low voltage networks: enhancing load visibility for network planning

Integrating electric vehicles (EVs) into low voltage (LV) networks introduces non-linear loads to the distribution grid, compromising grid stability. This study applies machine learning algorithms to detect EV charging events within LV grids, to enhance load visibility for network planning. By analysing voltage fluctuations, current spikes, and power quality deviations, the research aims to identify potential overloads caused by EV charging. The study uses data collected over 25 weeks from three-phase transformers at an office area charging station. The outcome is expected to enhance grid resilience, mitigate equipment damage, and ensure the sustainable integration of EVs into LV grids, supporting network planners in grid upgrades and smart technology integration. Results demonstrate that EV charging events can be detected using a linear Support Vector Machine (SVM) model with about 86% sensitivity and 78% specificity. Despite its proficiency in identifying EV charging events, the model seems to detect load fluctuations other than the EV-produced ones (false positives), evidencing the problem of load decoupling. Additionally, the presence of photovoltaic (PV) activity also seems to hinder model performance. Future efforts will focus on reducing false positives to improve EV detection performance and efficiently handle the presence of PV activity in the grid.