【摘要】 文章首先对主流的边缘AI设备进行了调查和分析，发现不同类型的板卡在执行并行推理任务时，各个设备的并发适配性不尽相同，而且计算效率会因为任务的复杂度而有所降低，为了解决这个问题，分析了主流的任务调度算法，架构了更加高效的任务调度系统EAC2S,提出了基于FAC的强化学习的任务调度算法，使用实际运行得到的经验作为训练样本，感知各异构设备间的差异，从而找到最优调度策略，并设计了实验验证了优化的有效性。更多还原

【Abstract】 This paper first investigated and analyzed the mainstream edge AI devices, and found that when different types of boards perform parallel inference tasks, the concurrent adaptation of each device is not the same, and the computational efficiency will be reduced due to the complexity of the task.In order to solve this problem, the mainstream task scheduling algorithms are analyzed.A more efficient task scheduling system EAC2S is constructed, and a task scheduling algorithm based on FAC reinforcement learning is proposed.The experience obtained from actual operation is used as training samples to perceive the differences among heterogeneous devices and find the optimal scheduling strategy.Experiments are designed to verify the effectiveness of the optimization.更多还原