Solving electric vehicle routing problem with recharging and battery swapping using a collaborative decision attention network

Abstract

With the growing demand for electric vehicles (EVs) in logistics and transportation, long charging times and limited driving range have emerged as significant challenges. Battery swapping offers a faster alternative to conventional charging, reducing downtime but introducing additional costs. This study investigates the electric vehicle routing problem with recharging and battery swapping (EVRP-RBS), which requires balancing range constraints with cost-efficiency. To address this, we propose a collaborative decision attention network (CDAN) based on deep reinforcement learning (DRL). CDAN jointly optimizes routing and charging strategies by training an encoder-decoder structured policy network. The EVRP-RBS is formulated as a two-action Markov decision process. The encoder extracts features from customer nodes, charging stations, and the depot, embedding them separately into a high-dimensional space. A self-attention mechanism is employed to capture the internode relationships, producing a global representation for downstream decision-making tasks. To effectively coordinate route planning and energy replenishment, we introduce a dual-attention decoder, which integrates two specialized attention modules—one for routing decisions and another for charging or battery swapping decisions. This architecture enables efficient integration of routing and charging considerations, significantly enhancing solution quality. Extensive experiments demonstrate that CDAN achieves competitive performance compared to both traditional and DRL-based baselines while exhibiting strong generalizability. Notably, the charging decision module plays a critical role in improving the overall solution quality.