Serrano-Lujan, LucíaToledo, CarlosColmenar, Jose ManuelAbad, JoseUrbina, Antonio2023-11-282023-11-282022-06-01Serrano-Luján, L., Toledo, C., Colmenar, J. M., Abad, J., & Urbina, A. (2022). Accurate thermal prediction model for building-integrated photovoltaics systems using guided artificial intelligence algorithms. Applied Energy, 315, 119015.0306-2619https://hdl.handle.net/10115/26611En este estudio de aplican técnicas de aprendizaje computacional (inteligencia artificial), en concreto una combinación de gramáticas evolutivas (GE) y evolución diferencial (DE). Estas técnicas se aplican a un conjunto extenso de datos para obtener modelos. Un sistema de monitorización almacenó información de temperatura, humedad y condiciones atmosféricas de un cubo ensamblado con módulos solares fotovoltaicos de silicio monocristalino cada cinco minutos. A partir de los datos, y utilizando GE+DE, se obtienen modelos que permiten estimar la temperatura del módulo fotovoltaico. Este dato es clave para realizar estimaciones sobre el potencial de generación eléctrica de la configuración, puesto que las altas temperaturas implican una bajada importante de rendimiento. Finalmente, se compara la precisión de los modelos obtenidos con el más representativo disponible en la literatura, llamado Sandia. Nuestros modelos consiguen mejorar la precisión en un 11%.Progress in development of building-integrated photovoltaic systems is still hindered by the complexity of the physics and materials properties of the photovoltaic (PV) modules and its effect on the thermal behavior of the building. This affects not only the energy generation, as its active function and linked to economic feasibility, but also the thermal insulation of the building as part of the structure’s skin. Traditional modeling methods currently presents limitations, including the fact that they do not account for material thermal inertia and that the proposed semi-empirical coefficients do not define all types of technologies, mounting configuration, or climatic conditions. This article presents an artificial intelligence-based approach for predicting the temperature of a poly-crystalline silicon PV module based on local outdoor weather conditions (ambient temperature, solar irradiation, relative outdoor humidity and wind speed) and indoor comfort parameters (indoor temperature and indoor relative humidity) as inputs. A combination of two algorithms (Grammatical Evolution and Differential Evolution) guides to the creation of a customized expression based on the Sandia model. Different data-sets for a fully integrated PV system were tested to demonstrate its performance on three different types of days: sunny, cloudy and diffuse, showing relative errors of less than 4% in all cases and including night time. In comparison to Sandia model, this method reduces the error by up to 11% in conditions of variability of sky over short time intervals (cloudy days).engBIPVPV module temperatureModule temperature estimationGrammatical evolutionDifferential evolutionMachine learningAccurate thermal prediction model for building-integrated photovoltaics systems using guided artificial intelligence algorithmsinfo:eu-repo/semantics/article10.1016/j.apenergy.2022.119015info:eu-repo/semantics/restrictedAccess