A DRL-Based Intelligent Task Offloading and Caching Strategy for IIoT in MEC

In IIoT (Industrial Internet of Things) systems, diverse smart devices generate various types of computational tasks, often with low latency requirements. Leveraging computation offloading techniques in MEC (Mobile Edge Computing), these tasks can be sent to MEC servers for auxiliary computing. However, the dynamic changes in the environment and the increasing number of smart devices in the system lead to competition for the limited resources on the servers, making it crucial to develop rational offloading strategies. Motivated by this, to ensure efficient and stable operation of IIoT systems, we design intelligent offloading and task caching strategies for single-edge mobile networks to reduce the overall execution latency of intelligent tasks. This paper establishes a system model from three perspectives: task offloading, task caching, and task queue scheduling, and formulates a joint optimization problem for task offloading and task caching. We transform the interaction process of the system into a Markov decision process (MDP) and propose a low-latency scheduling strategy based on DRL (Deep Reinforcement Learning). At each time slot, this strategy collects dynamic environment information and reference task priority values to make offloading and caching decisions. Simulation results demonstrate that the proposed algorithm achieves rapid and stable convergence and effectively reduces the completion latency of intelligent tasks.