An Adaptive Multi-granularity Dynamic Task Reallocation Method Based on CBBA

Aiming at the bottleneck of collaborative efficiency caused by multi-source heterogeneous dynamic disturbances (such as task addition, deletion/change, agent failure/online) in post-disaster rescue scenarios, this paper proposes an improved consensus bundling algorithm framework (CBBA-AMG) that integrates adaptive multi-granularity response and load balancing optimization. By constructing a time-resource composite load model and introducing a load balancing penalty term based on Jensen-Shannon divergence, the problem of uneven task distribution in the traditional CBBA algorithm in heterogeneous multi-agent systems is effectively alleviated. At the same time, an event-driven dynamic reallocation mechanism is designed to dynamically trigger differentiated response granularities (global reset/cluster collaboration/individual adjustment) based on the disturbance impact metric function I(ϵ) to minimize the impact of replanning on system stability. Simulation results show that the proposed load balancing mechanism reduces the Gini coefficient of task allocation by 3.9%, and the maximum-minimum load difference is reduced by 16%; the adaptive multi-granularity strategy is 6.9 times faster than the global reallocation response speed, and the cumulative task revenue is increased by 12.7%, significantly improving the task completion rate and robustness of the system in a dynamic environment.