Prediction of Daily Solar Irradiation Using CNN and LSTM Networks

Vakitbilir N., Hilal A., Direkoğlu C.

14th International Conference on Applications of Fuzzy Systems, Soft Computing, and Artificial Intelligence Tools, ICAFS 2020, Budva, Montenegro, 27 - 28 August 2020, vol.1306, pp.230-238 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume: 1306
  • Doi Number: 10.1007/978-3-030-64058-3_28
  • City: Budva
  • Country: Montenegro
  • Page Numbers: pp.230-238
  • Keywords: Deep learning, Long short-term memory network, One-dimensional convolutional neural networks, Solar irradiation forecasting
  • Middle East Technical University Affiliated: Yes


© 2021, The Author(s), under exclusive license to Springer Nature Switzerland AG.Greenhouse gas emissions from conventional energy sources are accelerating the global warming. To alleviate this issue, countries are focusing on renewable energy sources, especially solar enery, to meet the increasing energy demand. Forecasting solar irradiation on different time-horizons is crucial for the integration of the solar energy to the existing or future electricity grids. In this paper, we focus on solar irradiation prediction, and present performances of two different methods for Kalkanlı region of Cyprus. We use a one-dimensional Convolutional Neural Network (1D-CNN) and a Long-Short Term Memory netwok (LSTM) separately for prediction. In particular, 1D-CNN and LSTM networks are employed with two different time-series input datasets to predict one-day ahead global horizontal irradiation (GHI) for Cyprus. Performances of the networks are evaluated and compared.