A Hybrid Edge–Cloud IoT and Machine Learning Framework for Real-Time Reliability Assessment, Fault Detection and Efficiency Improvement in Distributed Photovoltaic Grids

The fast-growing widespread distributed photovoltaic (PV) generation has brought new challenges to the reliability, efficiency and fault tolerance in solar assets over wide areas.Traditional monitoring architectures tend to use the cloud-computing nodes as missions' central node that leads to latency, low level of situational awareness and lag in reacting to major events such as partial shading, inverter faultiness, sensor drift soiling and module degradation.To overcome these limitations, the present work presents a composite Edge-Cloud IoT and Machine Learning (ML)-based framework for real-time reliability evaluation, intelligent fault detection and efficiency optimization in distributed PV grids.The proposed architecture combines low-cost, edgecomputing nodes with cloud-based analytics to form a two-tier intelligence system, where fast real-time anomaly detection, data pre-processing and early-stage fault classification are conducted at the edge while deep learning "brainer" diagnostics (i.e., long-term prediction of degradation and fleet optimization) are achieved in the cloud.The framework is based on supervised and unsupervised ML models that are used for the detection of early performance spoilage, failure prediction as well as real-time estimates of PV yield at different time intervals under changing operational conditions (e.g.: Random Forests, LSTM networks, autoencoders and clustering algorithms).A multi-parameter reliability indicator is proposed based on operational, electrical and environmental parameters to monitor the health condition of the system.Tested on a distributed PV microgrid, the hybrid framework increases fault detection speed by 63%, decreases false alarms by 42% and finally leads to an increase in energy yield by as much as 6-11% due to predictive maintenance and optimized inverter setpoints.The edge-cloud coordination minimizes the communication cost and guarantees real-time response in transient events, while maintaining a long-term analytical capability for strategy decision.In summary, the proposed framework is an evolutionary contribution to PV monitoring by introducing a scalable intelligent adaptive solution for utility-scale commercial and residential PV distributed networks.The results shed light on how to build up more stable and economic, renewable energy gridintegrated infrastructures.