Aiming at the technical bottlenecks in the operation and maintenance of large-scale PV (photovoltaic) power stations, such as large amount of data, high response delay and idle edge computing power, this paper proposes a real-time operation and maintenance decision support system based on edge computing and AI. The system adopts "cloud-edge-end" three-level architecture, and completes data preprocessing and lightweight AI reasoning through edge nodes to realize millisecond fault response, The cloud is responsible for model training and strategy optimization to ensure the continuous learning and decision-making accuracy of the system. An early-warning model based on gated recurrent units (GRU) and a fusion physical model-based efficiency degradation assessment method are introduced to achieve early identification of faults such as hot spots and latent cracks, along with dynamic tracking of power generation performance. In the six-month actual measurement of 10MW power station in Ningxia, the F1 score of the system reaches 99.0%, the response time is less than 100ms, and the operation and maintenance cost is reduced by about 22%, which verifies its effectiveness in improving the intelligent operation and maintenance level of PV power station.
Fan et al. (Sun,) studied this question.