Data-Driven Optimization of Electric Vehicle Charging Stations

We consider data-driven power capacity allocation to the chargers of a Charging Station (CS) so that we can subsequently maximize the total value of future Electric Vehicle (EV) charging requests admitted and scheduled within their deadlines. We present an Integer Programming formulation and a constant-factor approximation algorithm for power capacity allocation to the CS's chargers, assuming that EV charging requests are known in advance. Given a power capacity allocation, for the realistic case where EVs arrive online, we present a competitive online admission and scheduling algorithm, which allows the CS to fully commit to a fixed charging rate and interval upon EV arrivals. Combining these algorithms with online learning, we present a data-driven approach to the combined problem of power capacity allocation and online EV scheduling and prove a logarithmic upper bound on its regret. Our approach exhibits strong theoretical guarantees and near optimal practical performance, as demonstrated by our experimental evaluation, without making any assumptions about future EV arrivals.