Adaptive Graph Learning for Joint Optimization of Speed and Quality in Supply Chains

Abstract

Balancing rapid delivery with strict quality compliance is a critical challenge in healthcare logistics. To address this, we propose an Adaptive Graph Learning framework centered on the dynamic joint optimization of speed and quality. Unlike traditional static models, our approach models the supply chain as an evolving, multi-relational graph. By ingesting heterogeneous operational data - such as delivery velocity and temperature volatility - into a unified representation, we introduce a Dynamic Multi-Modal Weighting mechanism. This mechanism uses reinforcement learning to automatically recalibrate the trade-off between temporal urgency and condition stability in response to real-time network disruptions. Furthermore, we employ graph attention networks to identify functionally robust supply communities, enabling targeted, facility-level optimizations. Extensive empirical evaluations on large-scale healthcare datasets validate our framework's efficacy. Notably, a real-world case study demonstrates a 37.8% reduction in emergency response time and a 68.3% reduction in quality defect rates. These results confirm that our adaptive approach significantly enhances supply chain resilience, providing a scalable, intelligent foundation for modern logistics.