Optimizing the Dual-Channel Drug Distribution Path of Medical Insurance Using Graph Attention Network

Abstract

Efficient path planning for dual-channel medical insurance drug distribution requires adaptive coordination among heterogeneous nodes under dynamic inventory, demand, and policy constraints. This study proposes a graph attention network (GAT)-based optimization framework that models hospitals, pharmacies, warehouses, and distribution hubs as a directed weighted graph and integrates multi-head attention with reinforcement learning to achieve intelligent path scheduling. The attention mechanism dynamically captures inter-node dependencies and learns context-aware representations, while the reinforcement learning scheduler continuously updates routing decisions according to realtime network states. Experimental evaluation demonstrates that the proposed approach reduces the average delivery time from 54.9 min to 37.3 min, increases the demand fulfillment rate to 92.1%, and lowers the overall operational cost by 29.8%, while significantly improving inventory coordination and response efficiency at high-demand nodes. The framework provides robust dynamic optimization capabilities for large-scale distribution networks and offers a transferable graph-based resource allocation strategy for intelligent communication infrastructures, where efficient information propagation, network topology optimization, and adaptive routing are critical to electromagnetic information transmission and distributed system management.