Low-Latency Communications for Community Resilience Microgrids: A Reinforcement Learning Approach

Machine learning and artificial intelligence (AI) techniques can play a key role in resource allocation and scheduler design in wireless networks that target applications with stringent QoS requirements, such as near real-time control of community resilience microgrids (CRMs). Specifically, for integrated control and communication of multiple CRMs, a large number of microgrid devices need to coexist with traditional mobile user equipments (UEs), which are usually served with self-organized and densified wireless networks with many small cell base stations (SBSs). In such cases, rapid propagation of messages becomes challenging. This calls for a design of efficient resource allocation and user scheduling for delay minimization. In this paper, we introduce a resource allocation algorithm, namely, delay minimization Q-learning (DMQ) scheme, which learns the efficient resource allocation for both the macro cell base stations (eNB) and the SBSs using reinforcement learning at each time-to-transmit interval (TTI). Comparison with the traditional proportional fairness (PF) algorithm and an optimization-based algorithm, namely distributed iterative resource allocation (DIRA) reveals that our scheme can achieve 66% and 33% less latency, respectively. Moreover, DMQ outperforms DIRA, and PF in terms of throughput while achieving the highest fairness.