Dynamic Logistics Path Optimization via Integrated Ant Colony Optimization and Reinforcement Learning

Abstract

Logistics path planning plays a critical role in improving the efficiency and cost-effectiveness of distribution systems, especially under dynamic traffic conditions. This paper proposes a hybrid path optimization model that combines Ant Colony Optimization (ACO) with Deep Reinforcement Learning (RL), specifically a Deep Q-Network (DQN), to address the limitations of traditional static planning algorithms. The model integrates real-time traffic conditions and historical logistics data into a dynamic directed graph structure. ACO is first used to generate high-quality initial paths, which are encoded to initialize the RL environment and guide early exploration. As the vehicle navigates, real-time traffic fluctuations such as congestion and road closures trigger immediate re-optimization via the RL agent and adaptive pheromone updates in ACO. The model is evaluated using a real-world logistics dataset with 30 customer nodes under time window constraints and varying dynamic scenarios. Experimental results demonstrate that the proposed method reduces average delivery route length from 56.9 km to 52.3 km and lowers fuel and operational costs by 27%, while also achieving 100% punctuality. These findings validate the model’s effectiveness, robustness, and potential for deployment in intelligent logistics distribution systems.