A machine learning-driven virtualized edge framework for real-time monitoring and anomaly detection in solar photovoltaics

The rapid development of solar photovoltaics requires the need to have monitoring systems because they require real time response and are also stable in a wide range of operating conditions. Typical SCADA systems are inexpensive although with limited temporal resolution and can only usually spot faults when performance has become noticeable. Machine learning solutions on clouds are more accurate but lack latency, connection reliance and are prone to privacy threats. This void creates the necessity of fine grained and real time anomaly detection. In this research, a virtualized edge architecture (VEAD) is suggested, which combines convolutional networks, the gated recurrent unit, and XGBoost classifiers into a containerized pipeline and deploys on a Raspberry Pi. The framework utilizes multi-sensor PV data to project the results into a VR-based digital twin by running the data locally. This article demonstrates that edge intelligence and immersive visualization can be used hand in hand without the need to hinder the other. The model offers quick calculations and results that are easy to decipher. This balance is checked by tests of VR-simulated faults. The detection accuracy was 96.2% and inference was under one second with an average of ~ 312 ms. There was lag in cloud baselines. Also, the edge response time was over 60% faster, which is an important difference when faults develop rapidly. Under shading, dust, and disturbances of inverter performance, performance remained constant.