A framework to investigate charger capacity utilization and network voltage profile through residential EV charging data clustering

A significant increase in the adoption of electric vehicles (EVs) is expected over the next decade. Hence, an investigation of the potential impact of EVs on the electricity grid is critical. This paper presents a framework for grid impact analysis of residential EVs using time series clustering (to perform sequential simulation) and discrete clustering (to estimate peak EV power consumption). This paper employs four clustering techniques, which are (i) K-means, (ii) hierarchical, (iii) DBSCAN, and (iv) fuzzy c-means, by analyzing 348 EV customers’ charging data. Clustering techniques have been implemented in Python, and power flow simulations have been performed using MATLAB/MATPOWER software. The results demonstrated daily and weekly EV profile clusters and EV charger utilization factors. The clustered EV profiles have been passed through to the power flow simulation to identify the network voltage violations. In the daily and weekly clusters, both K-means and hierarchical methods have two dominant clusters having 30 to 40% customers and two minor clusters with 10 to 20% customers. On the other hand, DBSCAN has one dominant cluster (in daily profile) with around 70% customers — as this method is used for anomaly detection. The fuzzy c-means has four almost similar clusters with around 25% customers. The abovementioned trend is evident in the network voltage violation heatmaps having more voltage violations in the weekdays evening (5:00 to 8:00 PM).