TY - JOUR
TI - Solar irradiation forecastby deep learning architectures
AU - Dagistanli Omer
AU - Erbay Hasan
AU - Yurttakal Ahmet Haşim
AU - Kör Hakan
JN - Thermal Science
PY - 2022
VL - 26
IS - 4
SP - 2895
EP - 2906
PT - Article
AB - Global solar irradiation data is a crucial component to measure solar energy  potential when we plan, size, and design solar photovoltaic fields. Often,  due to the absence of measuring equipment at meteorological stations, data  for the place of interest are not available. However, solar irradiation can  be estimated by ordinary meteorological data such as humidity, and air  temperature. Herein we propose two different deep learning methods, one  based on a deep neural network regression and the other based on  multivariate long short term memory unit networks, to estimate solar  irradiation at given locations. Validation criteria include mean absolute  error, mean squared error, and coefficient of determination (R2 value).  According to the simulation results, multivariate long short term memory  unit networks performs slightly better than deep neural network. Even  though both have very close R2 values, multivariate long short term memory's  R2 values are more consistent. The same is true for mean squared error and  mean absolute error.