Towards Experiments for Comparing Anytime Algorithms to Optimize Assignments of Electric Vehicles To Charging Stations

Abstract

Charging of electric vehicles (EVs) on highways must be efficient for ensuring e-mobility because of the comparably limited range of EVs and the potentially very long charging times during longer trips. To reduce the carbon footprint of EVs, they should be charged as much as possible with renewable energy, ideally when it is currently available. Hence, we use multi-objective optimization for the allocation of EVs to charging sites. Since the traffic situation is subject to rapid change, this needs to be dynamic ‘anytime’ optimization. In this paper, we address how to ensure that an algorithm used is really better in some aspect than another one. We propose controlled experiments in a simulation environment, where the resulting data are statistically tested to avoid believing into results that are due to random fuctuation. We both propose an experiment design and show how it was implemented for a specific comparison of such algorithms. In particular, we compare a genetic algorithm previously published with other algorithms and show its efficiency.