Solar power generation analysis and forecasting real-world data using LSTM and autoregressive CNN

tosun n., sert e., AYAZ E., YILMAZ e., GÖL M.

3rd International Conference on Smart Energy Systems and Technologies, SEST 2020, Virtual, Istanbul, Türkiye, 7 - 09 Eylül 2020 identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/sest48500.2020.9203124
  • Basıldığı Şehir: Virtual, Istanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: Artificial neural networks, Autoregressive convolutional neural networks, Long-short term memory, Solar power forecasting
  • Orta Doğu Teknik Üniversitesi Adresli: Evet

Özet

© 2020 IEEE.Generated power of a solar panel is volatile and susceptible to environmental conditions. In this study, we have analyzed variables affecting the generated power of a 17.5 kW real-world solar power plant with respect to five independent variables over the generated power: irradiance, time of measurement, panel's temperature, ambient temperature and cloudiness of the weather at the time of measurement. After our analysis, we have trained three different models to predict intra-day solar power forecasts of the plant. Our models are able to predict future power output of the solar power plant with less than 10% RMSE without requiring additional sensor data, e.g. a camera to observe clouds. Based on our forecasting accuracy, our study promises: fast, scaleable and effective solutions to solar power plant maintainers and may facilitate grid safety on a large scale.