Collaborative optimisation of multi-agent reinforcement learning in enterprise digital supply chain

With the deepening of the digital transformation of enterprise supply chain, multi-agent system collaborative optimisation faces core challenges such as low communication efficiency and strategy instability.This paper innovatively introduces cognitive load theory into this field, constructs a quantitative model of collaborative cognitive load, and proposes a collaborative optimisation framework based on hierarchical attention communication and distributed policy distillation.The framework minimises the external load through structured communication, and improves the utility of associated load through policy distillation to achieve efficient collaboration.Experiments based on Amazon's real supply chain data show that the proposed method significantly outperforms the current optimal baseline method in terms of order fulfilment rate (increased to 94.8%) and total logistics cost (decreased by 8.3%), and statistical tests confirm the significance of its performance improvement.This study provides a new theoretical perspective and practical tool for distributed intelligent collaboration.