Behavioral modeling of corporate charging hubs introducing a novel clustering method

As commercial vehicle fleets shift from combustion-based vehicles to Electric Vehicles (EVs), charging demand at corporate grid connection points increases, leading to higher peak loads and electricity costs. Concurrently, EVs at workplaces offer a great potential for increased own-consumption of renewables, demand–response programs, and ancillary grid services. The implementation of smart Energy Management Systems (EMSs) presents the opportunity to leverage these potentials while concurrently addressing and mitigating any associated drawbacks of EV charging. The design of algorithms necessitates a comprehensive understanding of the underlying charging behavior. We address this by evaluating Charge Detail Records (CDRs) of 419 AC charging and 50 DC charging Corporate Charging Hubs (CCHs) and segmenting them into representative behavior clusters. To this end, we are introducing a tree-based time-series clustering approach using Dynamic Time Warping (DTW). A total of seven representative AC charging clusters are identified, distinguished by their distinct charging activity patterns on weekends, daily power peak, and their flexibility potential. In a subsequent step, a conditional, physics-consistent, Gaussian Mixture Model (GMM)-based generator model is trained to capture the interdependencies of the underlying charging behavior patterns. The trained generator effectively emulates the original data distributions of the three dimensions of a CDR: start time, delivered energy, and session duration. This is evidenced by the high congruence between the generated and original data, as demonstrated by low D-values in a Kolmogorov–Smirnov (KS) test for all three dimensions (0.014-0.113). The utilization of the generator is conducive to the simulation of charging behavior within the context of EMS-design and helps to accelerate the development of smart charging strategies for a large, representative customer and user base.