Study Reveals Critical Efficiency and Transparency Gap in Microgrid Artificial Intelligence

A systematic bibliometric review of two thousand one hundred and fifty scholarly records indexed in the Scopus database between 2011 and 2025 has identified a significant structural dichotomy in the development of artificial intelligence for microgrid energy management systems. Published by a research team including Serge Raoul Dzonde Naoussi, François Dieudonné Mengue, David Tsuanyo, and Bernard Kamsu Foguem, the investigation maps the historical convergence and divergence of Explainable Artificial Intelligence and Frugal Artificial Intelligence within decentralized energy management. The authors utilized VOSviewer and the Bibliometrix R package to evaluate literature spanning fifteen years, uncovering that while research into algorithmic interpretability is undergoing exponential expansion driven by regulatory scrutiny and operational trust requirements, computational frugality remains a comparatively peripheral concern. Model designers rarely prioritize computational efficiency, memory conservation, or execution energy despite the stringent resource boundaries of physical edge deployments. The analysis demonstrates that contemporary engineering literature consistently overlooks the operational energy footprint generated by the optimization algorithms themselves.

Microgrids represent localized electrical distribution networks that coordinate distributed energy generation, including intermittent renewable sources such as photovoltaic panels and wind turbines, with localized energy storage devices and fluctuating consumer loads. Across contemporary power engineering, microgrids provide localized energy autonomy, bolster electrical network resilience against extreme weather events and grid disturbances, and facilitate community energy sharing while electrifying remote geographic regions where main utility grid extensions remain technically infeasible or cost-prohibitive. Within these networks, the Energy Management System serves as the supervisory decision layer, continuously monitoring volatile operational states and orchestrating coordinated control actions across power generation, battery storage, and dynamic consumption profiles. Traditionally guided by deterministic mathematical formulations, energy management systems have rapidly adopted machine learning and deep learning methodologies to resolve stochastic operational challenges, accommodate nonlinear battery degradation behaviors, and generate short-term weather and load forecasts when physical distribution infrastructure is partially unobserved or computationally challenging to model accurately.

This aggressive transition toward deep and heavily parameterized neural network architectures has introduced what the study characterizes as an operational complexity trap and an algorithmic energy paradox. Although highly parameterized deep neural networks achieve superior predictive accuracy and operational control capabilities, they are inherently data-intensive and computationally demanding, requiring substantial processing cycles, extensive memory bandwidth, and elevated electricity consumption during training and inference. In the context of microgrid infrastructure, this dynamic creates an ironic operational paradox: energy-intensive computing algorithms are deployed to maximize electrical efficiency, yet their computational demands frequently exceed the hardware capacities present in remote, bandwidth-constrained, or economically limited microgrid environments. Field-deployed energy management systems must satisfy real-time latency thresholds, maintain uninterrupted reliability under degraded network connectivity, and operate continuously on compact microcontrollers or localized computing units. Consequently, the escalating hardware and energy costs of complex models threaten practical implementation in decentralized utility networks.

To resolve these operational tensions, the researchers evaluate the concurrent integration of Frugal Artificial Intelligence and Explainable Artificial Intelligence as mutually essential, non-negotiable prerequisites for deployable energy management systems. Within the scope of the study, Frugal Artificial Intelligence designates modeling architectures and data processing pipelines designed specifically to minimize resource consumption across computational processing, memory footprint, data volume, and algorithmic electricity consumption while preserving high operational performance. Concurrently, Explainable Artificial Intelligence supplies interpretable rationales, transparent surrogate decision rules, and feature attributions for safety-critical microgrid management decisions, including automated load shedding, network islanding, and battery charge scheduling. Because automated control actions directly impact expensive electrical assets subject to non-trivial failure modes and physical degradation, utility operators and regulatory authorities cannot safely adopt opaque black-box formulations without verifiable auditability and clear pathways for human operator intervention.

The bibliometric investigation addressed four primary research questions examining publication evolution, actor geography, intellectual and conceptual structures, and critical literature gaps at the intersection of frugal computation and explainability. To ensure reproducibility and methodological transparency, the authors extracted a final corpus of two thousand one hundred and fifty publications from Scopus by applying a strict, multi-term Boolean retrieval strategy spanning the years 2011 through 2025. This deliberate query design required the simultaneous presence of microgrid energy management terminology, explainability descriptors, and computational frugality or green computing keywords. By demanding the explicit co-occurrence of all three topical domains, the researchers eliminated subjective thematic assumptions and focused directly on literature where authors self-identified their work within this conceptual intersection. The resulting dataset reflects the degree to which the international engineering community explicitly treats algorithmic efficiency and model transparency as coupled engineering objectives rather than isolated research tracks.

Standardized pre-processing procedures and analytical thresholds were configured within the VOSviewer and Bibliometrix software environments to stabilize generated network visualizations and eliminate metadata distortion. Author variants and institutional naming formats underwent systematic harmonization. Furthermore, an author-network filtering rule was established to exclude multi-authored publications containing more than eight contributors, mitigating artificial inflation of co-authorship ties driven by broad institutional consortia or review panels. Network mapping parameters incorporated full counting methodologies across multiple analytical units: author co-authorship was evaluated at a threshold of fifteen publications, institutional collaboration was mapped at four documents, and country-level cooperation captured jurisdictions with at least twenty publications. Direct document citation analysis isolated the top twenty-five highly cited papers meeting a threshold of three hundred citations, identifying twelve interconnected publications. Keyword co-occurrence stability was systematically assessed through a parameter sensitivity sweep evaluating thresholds at five, ten, twenty-five, fifty, seventy-five, and one hundred occurrences.

Descriptive production metrics demonstrate that research linking microgrid energy management with both computational frugality and explainability is primarily a recent academic development, exhibiting marked escalation between 2018 and 2025. This temporal intensification reflects a transition from early exploratory inquiries to mature deployment-oriented engineering problems. Dissemination tracking indicates a dual-channel publication pattern where rapid-cycle conference proceedings introduce novel algorithmic concepts and optimization formulations, while archival scholarly journals consolidate empirical validation, long-term stability assessments, and reproducibility standards necessary for critical utility infrastructure. Applying Bradford’s law revealed a concentrated core zone of publication venues that host a disproportionate share of the corpus, distinguishing outlets that structurally anchor knowledge accumulation from those that are only intermittently present. However, the authors observe that dominant dissemination channels rarely enforce systematic reporting standards for deployment indicators, leaving computational energy consumption, execution latency, memory demands, and practitioner-validated interpretability protocols largely unmeasured across published studies.

Geographic and institutional analyses show that scientific production is internationally distributed across diverse continents, yet collaborative networks depend heavily on a limited number of central cross-country hubs. Evaluating single-country publications alongside multiple-country publications demonstrates varying rates of internationalization, with high multiple-country publication shares functioning as an operational proxy for international knowledge exchange and the diffusion of shared benchmarking practices. Citation impact aggregated across organizations reveals that leading research institutions function as influential paradigm setters whose modeling preferences propagate widely through collaborative academic networks. When dominant research institutions prioritize algorithmic explainability while neglecting empirical measurements of computational energy or hardware constraints, this thematic bias diffuses across collaborating centers and downstream publication venues. The researchers argue that bridging this structural divide requires international collaboration consortia to establish shared benchmarks that balance transparency metrics with rigorous hardware profiling.

The conceptual architecture of the domain was reconstructed using keyword co-occurrence mapping, frequency treemaps, and Multiple Correspondence Analysis to project high-dimensional keyword associations into a low-dimensional conceptual space where proximity reflects thematic alignment. Across all analytical representations and threshold sensitivity sweeps spanning five to one hundred occurrences, interpretability terms consistently occupied the central, highly connected spine of the network, whereas frugality, edge execution, and green computing descriptors remained peripheral and fragmented. This structural disparity occurred despite the strict Boolean query requiring all captured papers to touch upon frugality terminology. Document-level citation networks and source-level citation mappings mirrored this divide: while citation links formed cohesive intellectual clusters around explainability frameworks and algorithmic forecasting architectures, papers addressing computational efficiency exhibited weaker connectivity, indicating that methodological norms surrounding explainability have solidified much faster than standardized practices for evaluating computational resource footprints.

Thematic maturity evaluations combining bibliographic coupling clusters and strategic centrality-density diagrams provide empirical justification for realigning future research priorities. Within the strategic diagram, which classifies topics into motor, basic, niche, and emerging or declining quadrants based on external interaction and internal cohesion, explainability-oriented themes occupy mature motor theme positions characterized by high centrality and density. Conversely, deployment-relevant themes such as hardware constraints, edge feasibility, and algorithmic maintainability remain situated in underdeveloped or emerging quadrants. The authors emphasize that this thematic divergence generates severe practical risks for microgrid modernization: highly interpretable models risk remaining unusable in real-world environments due to excessive compute requirements or reliance on external cloud servers, whereas compact, lightweight models may achieve edge deployment but lack the transparent reasoning necessary for critical utility decision-making. Overcoming this divide necessitates redefining state-of-the-art benchmarks away from pure predictive accuracy toward multi-objective deployment optimality.

As an essential pathway toward scalable deployment, the study highlights data-centric frugality, which shifts computational efficiency efforts upstream through data purification, noise filtering, outlier detection, sensor validation, and feature minimization alongside model compression techniques like pruning and quantization. In resource-constrained microgrids characterized by noisy telemetry and intermittent communications, reducing input dimensionality substantially decreases model training and inference burdens without sacrificing control reliability. Data-centric reductions naturally enhance interpretability by clarifying the specific physical signals governing energy management actions. Triangulating thematic clusters and Bradford source distributions, the authors outline an actionable research agenda for 2026 through 2030 focused on edge-ready explainable architectures with guaranteed inference latencies, robust operator utility, and mandatory reporting standards. This proposed disclosure standard requires future energy management literature to explicitly document physical hardware platforms, memory consumption, execution latency, and communication costs alongside conventional forecasting metrics.

Subject of Research: Bibliometric analysis of frugal and explainable artificial intelligence in microgrid energy management

Article Title: Frugal and explainable artificial intelligence for microgrid energy management systems: a bibliometric view

Article References: Dzonde Naoussi, S. R., Mengue, F. D., Tsuanyo, D., & Kamsu Foguem, B. (2026). Frugal and explainable artificial intelligence for microgrid energy management systems: a bibliometric view. Energy Informatics, 9(1), Article 61. https://doi.org/10.1186/s42162-026-00654-5

Image Credits: AI Generated

DOI: 10.1186/s42162-026-00654-5

Keywords: Microgrid Energy Management, Explainable AI, Frugal AI, Bibliometric Analysis, Power Systems, Edge Computing, Decentralized Energy, Energy Efficiency, Frugal, explainable, artificial, intelligence