DEVELOPMENT OF AN ADAPTIVE ROUTING ALGORITHM FOR C2C LOGISTICS USING REINFORCEMENT LEARNING

DOI: 10.31673/2412-4338.2025.017718

Abstract

This paper proposes an adaptive routing algorithm for C2C logistics based on the integration of demand forecasting using Long Short-Term Memory (LSTM) networks and Reinforcement Learning (RL). The algorithm is designed to optimize courier routes in dynamic urban environments, where traditional methods lack flexibility and efficiency. The proposed approach reduces route lengths, improves adaptability to demand fluctuations, and optimizes transportation costs. The algorithm consists of three key components: demand forecasting, environment state updating, and RL-based route correction. Demand forecasting is performed using an LSTM neural network, which allows for early identification of high-order density areas and minimizes delays. The RL component ensures real-time route adaptation, enabling optimal decision-making based on traffic conditions and the logistics network. Route correction incorporates predicted demand values, Q-function evaluations, and an optimal decisionmaking policy, enhancing resource utilization efficiency. The algorithm’s performance was evaluated in a simulation environment replicating real urban logistics conditions. A comparative analysis with traditional algorithms (heuristic methods and VRP) showed that the adaptive approach reduces the average route length by 3.3%-9.2%, depending on the number of recalculations. Delivery time was reduced by 21%, while overall costs decreased by 12–17%. However, the adaptive algorithm requires more computational resources: the average route computation time is 2–3 times higher than that of traditional approaches, and the number of recalculations may exceed the rational threshold of 15, necessitating further optimization. The results confirm that the proposed algorithm is effective in highly dynamic and variable demand scenarios, though its computational requirements may pose challenges for practical implementation. Future research should focus on improving route recalculation algorithms, implementing multi-agent systems for courier coordination, and integrating cloud computing to enhance overall performance.

Keywords: C2C logistics, adaptive routing, reinforcement learning, demand forecasting, multi-agent systems, last-mile logistics, route optimization, digital twins, cloud computing.