User preference cluster-based optimization for electric vehicles charging in smart buildings under uncertainty

This paper explores the use of data-driven techniques, comprehensively employing clustering methods, in smart charging applications within smart buildings to address uncertainties arising both from stochastic electric vehicle (EV) user behavior and from building load and photovoltaic (PV) generation. Through the analysis of two real-world datasets, the study aims to enhance the accuracy of user preference estimation in prioritized EV charging. Building upon an improved estimated preference, two cluster-based robust charging algorithms are introduced to better schedule EV charging. The K-means cluster-based current-window prioritized charging algorithm is tailored for scenarios with limited user preference information, while the K-means based consensus clustering (KCC) cluster-based scroll-window algorithm is optimized for scenarios with more user preference information complemented by historical user data. Additionally, the research examines the management of uncertainty in building load and PV generation using information gap decision theory (IGDT), presenting robustness cost analyses. Simulations in real-world office-building scenarios showcase the efficacy of the proposed algorithms. The cluster-based charging schedules demonstrate a close alignment with actual measured charging parameters, achieving daily operational cost reductions between 6.6% and 14.7% compared to charging schedules that do not incorporate prioritized charging. When scaling to 100 EV charging sessions, the proposed method achieves cost reductions of 35% to 40% with computational time increasing from around 0.5s to 4s, confirming its suitability for larger-scale deployments.