How to deploy battery swapping stations for electric vehicles

Abstract

The rapid adoption of electric vehicles (EVs) raises new challenges for the strategic planning of urban battery swapping station (BSS) networks, where infrastructure decisions and drivers’ responses are tightly interdependent. This paper develops a bilevel optimization model for BSS network design. The upper level represents the service provider’s decisions on station siting and operational capacity allocation, while the lower-level captures drivers’ travel and swapping choices induced by the resulting network configuration. This structure explicitly links provider decisions to user behavior, allowing swapping demand and station utilization to be determined endogenously rather than assumed exogenous. To solve the resulting large-scale problem, we propose a reinforcement learning–enhanced column generation algorithm (RL-CG) with two key innovations: (i) integrating reinforcement learning into the column generation framework to solve pricing subproblems more efficiently, and (ii) incorporating a multi-head attention mechanism to improve learning efficiency and scalability. Computational experiments show that RL-CG achieves the same solution quality as commercial solvers on benchmark instances. Meanwhile, it substantially reduces computation time as the problem size increases. Further sensitivity analyses yield actionable managerial insights: (i) urban structure strongly shapes spatial demand patterns and station utilization, implying that planning strategies should be tailored to city-specific mobility characteristics; (ii) range anxiety can affect network performance differently across urban contexts; and (iii) when upgrading charging stations to increase charging speed, prioritizing upgrades at a subset of key stations delivers larger operational gains than uniformly distributing upgrades. These results provide practical guidance for BSS operators and contribute new methodological tools for bilevel EV infrastructure planning.