Uso de aprendizado de máquina no desenvolvimento de modelos de previsão da temperatura de operação de células fotovoltaicas

Abstract

Designing a photovoltaic (PV) system involves estimating its electricity production, which is influenced by the environmental conditions and the PV cell operating temperature (Tc). Therefore, a model is needed to estimate Tc. This research aimed to develop a model for determining Tc based on Machine Learning (ML) techniques applied to a dataset of solar irradiation, ambient temperature, wind speed, cell temperature, day, year and hour collected at the Laboratory of Alternative Energy (LEA) of the Federal University of Ceará (UFC). Initially, a study of the recent models for determining Tc through a literature review is presented. 33 equations found in the literature to estimate Tc are summarized in just 3 general forms. Subsequently, the dataset collected between 2018 and 2020 in the LEA is treated and used in the application of regression learning techniques. The techniques applied are Linear Regression (RL), Support Vector Machine (SVM), Gaussian Process Regression (GPR) and Neural Network (NN). The results obtained by the application of RL, SVM, GPR, and NN are compared, such models presented the average magnitude of the error, the RMSE, of 3.5377; 2.8374; 2,651; and 2.9706 degrees in the test set, respectively. The best ML model developed, the GPR model, is compared with some conventional models present in the literature for forecasting Tc and presents the best performance between those.