In real-world logistics transportation, leveraging historical route data can provide valuable insights into drivers' route preferences, enabling them to avoid potential risks and enhance route planning reliability. Based on this, this paper studies the vehicle routing problem with the driver's experience, introduces a dual path evaluation index considering both the path reliability and driving distance, and then establishes the corresponding integer programming model. On the basis of fully analyzing the characteristics of the problem, a knowledge-based dynamic multi-start variable neighborhood search algorithm is proposed. Firstly, generalized sequence pattern mining techniques are employed to extract experience paths, including frequent and potential sequence, from a large dataset of vehicle trajectories. Then, a knowledge-based conflict resolution strategy is proposed to construct high-quality initial solutions by integrating the aforementioned experience paths. Finally, a dynamic multi-start variable neighborhood search algorithm is introduced to improve the initial solutions. Through empirical analysis using real logistics distribution data from a jewelry company, the proposed algorithm demonstrates significant improvements compared to traditional variable neighborhood search algorithms. It effectively reduces the scale and solving time of the problem, while simultaneously minimizing driving distance and improving the reliability of path planning, which provide a valuable decision-making foundation for path planning in actual logistics enterprises.