Transformer-Based Models for Streaming Anomaly Detection in Solar Microgrids: A Systematic Review

Abstract Background: Solar Photovoltaic (PV) microgrids operate in highly non‑stationary conditions, where irradiance, weather, and load variability cause frequent concept drift in streaming telemetry. Traditional anomaly detection models degrade under these dynamics. Objective: To systematically review Transformer-based models for streaming anomaly/fault detection in energy systems, emphasizing solar PV and microgrids and appraise evidence on (i) drift adaptation, (ii) edge deployment efficiency, and (iii) interpretability for operator use. Methods: We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines. Records published between January 2023, and May 2026 were screened reflecting the recent emergence of Transformer-based anomaly detection research addressing streaming operation, concept drift, and edge deployment. Inclusion required a Transformer/attention-based architecture applied to anomaly/fault detection on time-series or operational telemetry, with rule-based removal of clearly irrelevant studies followed by manual validation. Results: A total of 2968 records were identified. After duplicate removal and screening, 113 studies were included in the final synthesis. Transformer-based architecture is widely reported across included studies but shows uneven application across domains. Photovoltaic systems account for most studies, while microgrid-specific applications remain limited. Reporting of drift adaptation, edge efficiency, and interpretability is inconsistent. Conclusions: Transformers show strong anomaly detection potential, yet gaps persist in drift-aware training, standardized edge metrics, and validated interpretability. We recommend lightweight attention mechanisms, integrated drift detectors, and operator-tested explanations to enable reliable PV microgrid deployment.