Artificial Intelligence for Predictive Generation and Maintenance of Photovoltaic Power Plants
Abstract
The integration of Artificial Intelligence (AI) into the operational management of photovoltaic plants promotes significant advances in generation forecasting, predictive maintenance, and performance optimization. This study uses data from 540 inverters of 250 kW and 54 solarimetric stations distributed across five Brazilian states, collected through the IsolarCloud platform and processed within the InterSystems IRIS Data Fabric environment. In IRIS, the data are integrated into a scalable Data Lake that consolidates both structured and unstructured information, ensuring governance, traceability, and interoperability. This infrastructure supports supervised and unsupervised AI models, as well as vector search mechanisms based on embeddings and Hierarchical Navigable Small World (HNSW) indexing, enabling semantic correlation and explainability. The results show an average Mean Absolute Percentage Error (MAPE) of up to 8%, 92% accuracy for 24-hour and 7-day horizons, a reduction of up to 38% in OPEX/MWp, an increase of 27% in Mean Time To Failure (MTTF), and a 32% reduction in Mean Time To Diagnose (MTTD). It is concluded that the combination of AI, semantic vector search, and integrative data architecture establishes a new paradigm of explainable operational intelligence applied to solar energy, enhancing the predictability, reliability, and efficiency of large-scale photovoltaic plants.
Keywords
solar energy, photovoltaic forecasting, artificial intelligence, predictive maintenance, Data Fabric, Vector Search