Adaptive Edge-Computing Framework for Real-Time Fault Detection and Performance Optimization in Smart Grid Photovoltaic Systems

The rapid decentralization of modern energy grids demands resilient, low-latency monitoring solutions to mitigate localized failures and optimize solar power generation. Conventional cloud-reliant supervisory systems frequently encounter bandwidth bottlenecks and unacceptable communication latencies during high-frequency telemetry events. This paper introduces an edge-native IoT architecture integrated with lightweight transformer-based anomaly detection algorithms designed for real-time diagnostics in grid-tied photovoltaic arrays. By embedding inference capabilities directly into distributed micro-inverter gateway devices, the system processes multi-sensor electrical, thermal, and irradiance telemetry locally with sub-second response times. Field validation conducted across an operational 5 MW microgrid testbed demonstrated an anomaly identification accuracy of 98.2% alongside an 84% reduction in uplink bandwidth consumption compared to standard cloud ingestion pipelines. The findings confirm that distributed edge intelligence significantly accelerates fault isolation, reduces operational downtime, and extends photovoltaic hardware lifecycle in large-scale renewable networks.