Abstract: Aiming at the delay-sensitive multi-dependent task scheduling problem of smart cities, this paper proposes a smart city architecture empowered by edge computing and designs a computation offloading method to meet the scheduling requirements of tasks. Firstly, this paper first establishes a multi-dependent task model, as well as a latency constraint for the task and a load balancing constraint model for the smart city server. Secondly, agents that perceive dependencies between tasks are trained using deep reinforcement learning algorithms to make computational transfer decisions in real-time. Finally, a series of experiments are conducted to verify the effectiveness of this method in latency and energy consumption optimization.

Key words: smart cities, computation offloading, latency-sensitive task, multi-dependency task, deep reinforcement learning