Internal seminar: Claire Fragkedaki

The current work presents a novel approach to the (Decoupled) Electric Vehicle Routing Problem. The primary objective is to reduce charging-related delays and optimize the utilization of electric tractor fleets by decoupling trailer and tractor components, thus improving logistics efficiency. The proposed model utilizes a Deep Reinforcement Learning algorithm with Transformer Architecture to solve this problem. The study aims to optimize routing within a network of charging stations, by individually selecting tractors and trailers, as well as determining the sequence of charging station visits. Building upon previous work in related problems, such as the Traveling Salesman and the Heterogeneous Vehicle Routing Problems, the proposed framework contributes to the advancement of D-EVRP solutions. The current research offers practical insights for improving the efficiency of logistics operations involving electric tractors, paving the way for future applications in larger-scale operations and the integration of emerging technologies like autonomous vehicles. Experimental results based on randomly generated instances underscore the model’s potential. While the baseline outperforms the model, its inherent flexibility and dynamic decision-making provide a promising foundation for future advancements.