Clustering Based Method anomaly definition in Electric Vehicles charging curves

Predictive maintenance of Electric Vehicles charging stations is of crucial importance to reduce downtime, considering the rapid rise in electric vehicle usage and the consequent increasing electric load demand. Future faults on which maintenance is planned are predicted considering previous anomalous behavior in the electric load and previous faults, anomalies detection models are commonly used to check for these unpredictable behavior in the data. Classification of anomalies found by the detection model can help identify connections between the anomalies and consecutive faults. In this paper, we propose a clustering technique applied to anomalous charging session curves output from an anomaly detection model. The main feature on which hierarchical clustering is applied is the shape of segments of the anomalous sessions extracted through a sliding window technique. The metric used to compare the curves is Dynamic Time Warping. The model is validated on a real-world dataset of electric vehicle charging units energy readings.