Deep Reinforcement Learning-Based Service-Oriented Resource Allocation in Smart Grids

Resource allocation has a direct and profound impact on the performance of resource-limited smart grids with diversified services that need to be timely processed. In this paper, we investigate joint communication, computing, caching resource problem distinct delay requirement in grids. This paper aims optimize long-term system utility based reward loss function. Considering unknown dynamic environment as well huge state action space grids, deep reinforcement learning algorithm polling method is exploited learn policy by interacting environment. Specifically, edge nodes (ENs) act agents enable schedule resources appropriately. Then, are allocated service requirements queried according mechanism well-designed function utilized update strategy. Extensive simulation results show proposed outperforms three known baseline schemes terms network decision results. Besides, face large number still surpasses existing several schemes, especially improvement cache hit rate decrease computing delay.