Multi-Agent Meta-Reinforcement Learning for Self-Powered and Sustainable Edge Computing Systems

The stringent requirements of mobile edge computing (MEC) applications and functions fathom the high capacity dense deployment MEC hosts to upcoming wireless networks. However, operating such can significantly increase energy consumption. Thus, a base station (BS) unit act as self-powered BS. In this article, an effective dispatch mechanism for networks with capabilities is studied. First, two-stage linear stochastic programming problem formulated goal minimizing total consumption cost system while fulfilling demand. Second, semi-distributed data-driven solution proposed by developing novel multi-agent meta-reinforcement learning (MAMRL) framework solve problem. particular, each BS plays role local agent that explores Markovian behavior both generation transfers time-varying features meta-agent. Sequentially, meta-agent optimizes (i.e., exploits) decision accepting only observations from its own state information. Meanwhile, estimates policy applying learned parameters Finally, MAMRL benchmarked analyzing deterministic, asymmetric, environments in terms non-renewable usages, cost, accuracy. Experimental results show model reduce up 11% usage 22.4% (with 95.8% prediction accuracy), compared other baseline methods.