(Faculty of Engineering and Natural Science, 2024) Akhmetbek Y.
The Stochastic Dynamic Vehicle Routing Problem (SDVRP), in which customer demands are dynamic and follow a Poisson distribution, is examined in this study. We provide an SDVRP-specific Reinforcement Learning (RL) algorithm and evaluate it against algorithms for Random Selection, Largest-Demand Selection, and Max-Reachable Selection. We also use a Multi-head Attention architecture into our RL algorithm to better capture complex relationships in the dynamic routing environment. We show, by thorough evaluation, how RL with Multi-head Attention may be used to optimize resource usage and route efficiency, providing useful insights for addressing challenging logistics problems in real-world situations.