Deep Feature Selection of Meteorological Variables for LSTM-Based PV Power Forecasting in High-Dimensional Time-Series Data

• Published in: Algorithms Vol. 18; no. 8; p. 496
• Main Authors: Mauladdawilah, Husein, Balfaqih, Mohammed, Balfagih, Zain, Pegalajar, María del Carmen, Gago, Eulalia Jadraque
• Format: Journal Article
• Language: English
• Published: Basel MDPI AG 01.08.2025
• Subjects: Accuracy; Algorithms; Alternative energy sources; Artificial intelligence; Atmospheric moisture; Climate; Complexity; Correlation analysis; deep learning; Efficiency; Feature selection; Forecasting; Ground stations; Humidity; Hydrometeorology; International economic relations; Irradiance; long short-term memory; Machine learning; mean absolute; Meteorological data; Meteorological research; meteorological variables; Photovoltaic cells; Radiation; Research methodology; Solar energy; Solar energy industry; Solar farms; Solar power generation; Statistical methods; Variables; Weather stations; United Kingdom; New Jersey
• ISSN: 1999-4893; 1999-4893
• DOI: 10.3390/a18080496

Accurate photovoltaic (PV) power forecasting is essential for grid integration, particularly in maritime climates with dynamic weather patterns. This study addresses high-dimensional meteorological data challenges by systematically evaluating 32 variables across four categories (solar irradiance, temperature, atmospheric, hydrometeorological) for day-ahead PV forecasting using long short-term memory (LSTM) networks. Using six years of data from a 350 kWp solar farm in Scotland, we compare satellite-derived data and local weather station measurements. Surprisingly, downward thermal infrared flux—capturing persistent atmospheric moisture and cloud properties in maritime climates—emerged as the most influential predictor despite low correlation (1.93%). When paired with precipitation data, this two-variable combination achieved 99.81% R2, outperforming complex multi-variable models. Satellite data consistently surpassed ground measurements, with 9 of the top 10 predictors being satellite derived. Our approach reduces model complexity while improving forecasting accuracy, providing practical solutions for energy systems.