Data-Driven Low-Rank Tensor Approximation for Fast Grid Integration of Commercial EV Charging Stations Considering Demand Uncertainties

Abstract

The high power demand and charging variation of commercial fast electric vehicle (EV) charging stations has the potential to significantly impact the operation of electric power distribution systems. To evaluate this impact, the engineering practice is to run power flow studies at peak load, leading to a conservative result, or implement computationally expensive quasi-static time-series simulations. To overcome these drawbacks, this paper proposes a computational approach based on low-rank tensor approximation (LRA) for fast grid impact studies on distribution grid operation caused by commercial EV charging stations with demand uncertainties. To this end, a stochastic transportation model is implemented to predict the potential charging profiles of EV stations based on vehicles’ driving patterns, and the predicted charging demand is used to construct polynomial bases for the LRA algorithm. A methodology based on clustering approaches is proposed to strategically select the experimental samples for optimized determination of the coefficients, and the rank and degree of the LRA model are adaptively configured for improved performance. Test cases on a real-world utility feeder have demonstrated that the proposed approach is able to potentially reduce the time-series simulation time of distribution feeder circuits by around 99.99% while maintaining a high accuracy.