Research On Low-Latency And Energy-Efficient Task Decomposition And Scheduling Mechanisms For Intelligent Edge Computing

With the rapid development of Internet of Things(Io T)technology and the widespread application of intelligent devices,application scenarios such as autonomous driving and collaborative control of industrial equipment have increasingly higher requirements for latency and energy consumption.To meet the development needs of latency-sensitive and energysensitive applications,Mobile Edge Computing(MEC)is an effective solution to overcome the limitations of traditional cloud computing.MEC can fully utilize the computing resources of edge devices,improve computational efficiency,and reduce data transmission latency and terminal device energy consumption.Existing related research on computation offloading faces issues such as low efficiency in perceiving MEC network resources,excessive reliance on edge or cloud participation in endto-end collaborative computing,and the inability to fully utilize node computing resources due to unreasonable decomposition of computing tasks.This study addresses these issues by constructing a large and small resource tree for perceiving MEC network resources,enabling edge devices to perform collaborative computing,and employing multi-granularity task decomposition and adaptive priority adjustment to fully utilize MEC network resources.The specific contributions are as follows:(1)In response to the issues of computing devices being unable to perceive MEC network computing resources and task scheduling in complex environments,we propose a large and small resource tree edge discrete device resource management strategy and a task scheduling algorithm based on Double Deep Q-Network(DDQN).Nodes within the MEC network actively report resource information to the Edge Server(ES)to construct a small resource tree,and ES reports regional resources to the domain server to build a large resource tree.The DDQN algorithm then schedules terminal computing tasks to nodes in the resource tree with sufficient resources to complete computation offloading.Experimental results show that the proposed algorithm has lower latency and energy consumption,with average utility values 6.31%,8.13%,and 13.01% higher than those of pure DQN,DISCO,and full offloading task scheduling strategies,respectively.(2)To address the bottleneck problem of computing latency and energy consumption in edge environments that cannot be further reduced,we propose a resource-aware multigranularity task decomposition algorithm and a fuzzy logic-based adaptive priority decision algorithm.The Long Short-Term Memory(LSTM)network predicts the remaining node resources at future moments,performing multi-granularity decomposition of computing tasks to improve the matching degree between subtasks and MEC network resources.Edge servers recalculate priority by perceiving the urgency of tasks,adaptively allocating more computing resources to higher priority computing tasks.Experimental results show that the proposed algorithm has higher utility values and can further reduce the overall latency and energy consumption of computing tasks,with average utility values 1.13%,1.72%,and 2.26% higher than those of pure DQN,DDL-CORA,and full offloading strategies,respectively.The resource management and task scheduling strategies,multi-granularity task decomposition algorithm,and adaptive priority decision algorithm proposed in this study effectively reduce computing latency and energy consumption,having significant practical implications for the development of novel latency and energy-sensitive applications.