Application of a machine learning based optimization algorithm in the prediction of photovoltaic generation Aplicación de un algoritmo de optimización basado en aprendizaje automático en la predicción de la generación fotovoltaica

Abstract

Harnessing solar energy to generate photovoltaic electricity is crucial for sustainable development. This research proposes a photovoltaic generation prediction model using Machine Learning, specifically the “Random Forest” technique, based on real solar irradiation and temperature data for the year 2020. The data were obtained with a Solar Power Meter SM206 pyranometer in Lacatunga, Ecuador. The model was developed in Python, analyzing the most influential variables. It predicts power based on solar irradiance and temperature, with the month of May selected for having the largest amount of data, which allows keeping the error within acceptable margins and guarantees the efficiency of the model. It was compared with data from March, a month with less available data. The daily prediction accuracy was 91.41% and the monthly prediction accuracy was 94.47%. The monthly prediction is more effective due to the greater amount of data per variable, improving training performance. This model is reliable for predicting photovoltaic generation and optimizing solar energy systems, taking advantage of the high solar irradiation rates in tropical regions.