A joint machine learning and optimization approach for incremental expansion of electric vehicle charging infrastructure

Abstract

As Electric vehicle (EV) adoption increases worldwide, public charging infrastructure must be expanded to meet the growing charging demand. Furthermore, insufficient and improperly deployed public charging infrastructure poses a real risk of slowing EV adoption. The infrastructure thus needs to be expanded intelligently and flexibly while accounting for uncertain dynamics in future charging demand. Nevertheless, current methods for demand-based expansion often rely on rigid and error-prone tools, such as travel surveys and simple rules of thumb. The former is more appropriate for long-term, equilibrium scenarios, where we consider the charging network as a whole rather than incrementally expanding it. At the same time, the latter relies on business experience in a rapidly changing field. We propose a predictive optimization approach for intelligent incremental expansion of charging infrastructure. At each time step, we estimate the future charging demand through a Gaussian Process, which is subsequently used in a linear chance-constrained optimization method to expand the charging infrastructure incrementally. To develop and validate this framework, we account for environmental feedback by simulating user behavior changes based on historical charging records and considering an optimized charging network at every iteration. We apply this approach to a case study of EV charging in Dundee, Scotland. We compare different strategies and reasons for their pros and cons for monthly incremental expansion of the charging network. In particular, combining machine learning and optimization results in the cheapest expansion and one that serves the most demand.