User regret psychology-driven electric vehicle charging navigation strategy based on deep reinforcement learning and transfer learning

• A regret theory-based electric vehicle charging navigation model with multiple uncertainties is proposed. • Transfer learning-based method to estimate attributes of comparison strategies under uncertainty • Double deep Q-network enables fast and effective solving of the multi-uncertainty model. • Case studies show superior optimality, adaptability, and scalability of the proposed strategy. Current electric vehicle charging navigation (EVCN) models typically treat the problem as a static, one‐shot decision, neglecting the user regret psychology and myriad uncertainties that unfold throughout the charging journey. Moreover, by restricting decision‐making to a predefined roster of conventional charging strategies and simply selecting the option with the lowest retrospective regret, these methods risk overlooking novel or dynamic policies that could yield even greater satisfaction. To address these problems, a user regret psychology-driven EVCN strategy based on deep reinforcement learning (DRL) and transfer learning (TL) is proposed. In the EVCN problem, multiple uncertainties are considered, and several commonly used charging strategies are selected as comparison strategies to construct a regret-theory-based EVCN model. Subsequently, to accurately estimate the attributes of comparison strategies under uncertain conditions, a TL-based attribute estimation method is developed, utilizing the replay buffer and an improved Jaccard similarity. Based on this, a DRL algorithm is used for fast and efficient model optimization. Finally, the proposed method is simulated on the transportation network, and the results demonstrate that the proposed method improves the driving time estimation accuracy of comparison strategies, significantly lowers the charging strategy regret, and exhibits excellent adaptability and scalability.