Machine Learning and M2M Communication in Smart Grids: A Review, Taxonomy, and Future Directions for Fault Management

@article{1717009,
      author = {Onwughalu Markanthony Kenechi, Eseosa Omorogiuwa, Ehikhamenle Matthew},
      title = {Machine Learning and M2M Communication in Smart Grids: A Review, Taxonomy, and Future Directions for Fault Management},
      journal = {Iconic Research And Engineering Journals},
      year = {2026},
      volume = {9},
      number = {10},
      pages = {3684-3701},
      issn = {2456-8880},
      url = {https://www.irejournals.com/formatedpaper/1717009.pdf},
      abstract = {The transformation of power distribution systems toward smart grid architectures has intensified the need for intelligent, communication-aware, and resilient fault management solutions, particularly in developing-region networks where reliability indices remain critically constrained. While machine learning techniques have demonstrated transformative capabilities in fault detection and classification, achieving superior accuracy through deep graph learning, spatial-temporal recurrent neural networks, and hybrid artificial intelligence approaches, these studies predominantly operate under idealised communication assumptions that ignore the latency, jitter, and packet loss inherent in real-world machine-to-machine deployments. Conversely, existing machine-to-machine communication protocols for smart grids, including LoRaWAN, NB-IoT, and ZigBee, have been studied in isolation from the diagnostic algorithms they are intended to support. This critical disconnect between algorithm accuracy and deployment reality creates a significant gap in the literature: no unified framework currently exists to evaluate machine learning performance under realistic machine-to-machine communication constraints or to optimise communication parameters for diagnostic reliability.This paper presents a comprehensive review of machine learning and machine-to-machine communication integration for low-voltage and medium-voltage distribution fault management, structured around a novel taxonomy of communication-aware architectures. We systematically analyse existing approaches across four categories: communication-agnostic machine learning, communication-assisted diagnostics, communication-resilient algorithms, and fully integrated machine-to-machine machine learning systems. Through comparative analysis of verified literature spanning deep reinforcement learning for service restoration, multi-agent coordination for automated switching, and federated learning for distributed intelligence, we identify critical research gaps, including the absence of electrical-communication co-simulation platforms, underdeveloped edge-based inference architectures, and insufficient validation under non-independent and identically distributed data conditions. We further propose a unified conceptual framework integrating electrical feeder dynamics, machine-to-machine communication impairments, and machine learning inference within a coordinated architecture, validated against Nigerian distribution network parameters as a representative developing-region case study. By consolidating existing knowledge and highlighting the imperative for communication-machine learning co-design, this work provides clear directions for advancing intelligent, resilient, and deployable fault management systems in next-generation distribution networks.},
      keywords = {Machine-To-Machine Communication; Machine Learning; Fault Detection; Distribution Networks; Smart Grids; Communication-Aware Architectures; Co-Simulation; Developing Regions},
      month = {April},
      doi = {https://doi.org/10.64388/IREV9I10-1717009}
  }