Towards Using Machine Learning to Generatively Simulate EV Charging in Urban Areas

One of the major pushes to fight climate change is the decarbonization of energy and mobility, which are closely related and together significantly contribute to global carbon emissions[1]. Within this intersection, the electrification of mobility via large-scale deployment of electric vehicles (EVs) is one of the potential tools to decarbonize mobility in the coming decades[2, 3]. This push, however, requires significant investment in power infrastructure, mainly on the distribution level operated by distribution system operators (DSOs)[4, 5]. This is due to the need to expand the availability of not only private but also public charging infrastructure, which can help better distribute loads across space and time[6, 7]. Since power engineering infrastructure expansion is an effort requiring significant time and financial resources, a critical challenge in this area is how to optimize grid expansion for efficiency to cover anticipated EV charging demand in various areas while not overloading the current network and not overspending on areas where demand is not as high[8, 9]. This is especially difficult for DSOs since there has been a general lack of studies demonstrating analysis of real-world EV charging data in different geographies, mainly due to the data being vendor-locked and treated as confidential[10, 11, 8].