Dynamic Task Allocation for Edge-Cloud Systems Based on Hierarchical Reinforcement Learning

This paper addresses the task allocation problem for time-sensitive tasks on construction sites within the edge-cloud system framework. Task completion rate serves as a critical metric for evaluating allocation strategies. To enhance task completion rates, we propose a task completion model. Additionally, traditional reinforcement learning faces significant challenges when dealing with high-dimensional action outputs. We propose the HTD3 algorithm, leveraging hierarchical reinforcement learning and two neural networks to reduce action dimensionality. To avoid unreasonable outputs, we incorporate task urgency into the action execution process. Experimental results validate the effectiveness of our approach.