In the rapidly evolving field of logistics, the vehicle routing problem (VRP) and its dynamic counterpart, termed dynamic VRP (DVRP), have garnered significant attention from both industry and academia. Unlike classical VRPs, which conduct path planning based on a fixed problem definition, DVRPs require efficient and effective re-planning strategies to address real-time issues that arise from dynamic vehicle routing, such as new requests from unseen customers. Given the overlap between successive changes in DVRPs, valuable routing strategies can be derived from previous solutions to generate high-quality solutions for subsequent moments. In this paper, we introduce a novel approach to optimize DVRPs in response to unseen events by leveraging knowledge transfer from customer representations. To evaluate the performance of our proposed method, we conduct a comprehensive empirical study using 20 commonly used DVRP instances with diverse properties. The results from these instances demonstrate the effectiveness of the proposed algorithm compared to existing methods.