<p>This study investigates the application of Graph Neural Networks (GNN) for optimizing node selection and path planning within urban logistics distribution networks, aiming to address the challenges of cost efficiency, scalability, and adaptability under dynamic conditions. By modeling logistics networks as graph structures, the proposed framework integrates topology optimization and transportation scheduling into a unified approach, enabling more accurate and robust decision-making compared to traditional algorithms. Two publicly available datasets, the California Transportation Network (CTN) and OpenStreetMap (OSM), were employed to evaluate performance across diverse logistics scenarios, including varying traffic conditions, demand fluctuations, and different network scales. Experimental results demonstrate that GNN significantly reduces transportation costs, shortens delivery times, and enhances routing efficiency, while maintaining high computational scalability and convergence speed under multiple hardware configurations. Statistical testing confirmed the significance of these improvements, and computational resource specifications further validate transparency. Additional analyses, such as ablation studies and case-based validation, highlight the contributions of attention mechanisms and multi-task learning while confirming the framework’s practical relevance in real-world logistics operations. Overall, the findings provide strong evidence that GNN-based optimization can enhance the efficiency and resilience of urban logistics systems, offering theoretical advancement and practical guidance for intelligent logistics development.</p>