Forecasting Solar Power: Analysis of using a LSTM Neural Network

Increasing generation of renewable energies are good for the environment,
however bring challenges for electricity grid stabilization. Therefore, forecast-
ing of such energies becomes very important. This research conducts research
on forecasting solar power using a LSTM neural network. Data from BSRN
with a time frame of 1 min is used to built a model. Humidity, Dew/Frost
point, Station Pressure, month number and day number were found to work
best to predict solar radiation. Further the tuning, evaluation and testing of
the model are extensively discussed. An analysis of power spectral density
of the predicted values shows that there exist difficulties in predicting in low
frequencies. The LSTM model shows overall promising results and can be
used for various data sizes.

For data please request at BSRN: https://bsrn.awi.de/