Improved MAPPO-based Collaborative Computing Offloading Algorithm in Broadband Mobile Internet

Collaborative computation offloading technology can offload computational tasks from mobile devices to mobile edge servers and idle devices, thus breakthrough the performance bottleneck of MEC computation offloading systems and providing a new idea to meet the needs of computation-intensive mobile applications. In collaborative computation offloading systems, variations in the number of mobile devices cause changes of their state space and action space, and traditional deep reinforcement learning-based computation offloading decision algorithms (e.g., MADDPG) are usually unavailable at this case. A computational offloading algorithm based on improved MAPPO is proposed to transform variable dimensional input information into fixed dimensional aggregated information through the attention mechanism, ensuring that the input dimensions of the policy network and value network do not change with the number of smart devices, which meet the demand for flexibility and scalability of computational offloading decision algorithms in collaborative computational offloading systems.