SOLAR PHYSICS, vol.296, no.11, 2021 (SCI-Expanded)
The aim of this study is to predict solar activity for the next 10 years (Solar Cycle 25) using a deep learning technique known as a stateful Long Short-Term Memory (LSTM) network. To achieve this goal the number of daily sunspots observed by the American Association of Variable Star Observers (AAVSO) organization from 1945 to 2020 is used as training data for a stateful LSTM model. Time slices are produced by dividing the data between 1945-2020; then data are predicted and examined for the test years of the network trained on these time slices. The mean and smoothed values are calculated from the estimated daily data, compared with the actual mean, and smoothed values, including standard deviations, and the prediction accuracy of the model is examined. Finally, the number of daily sunspots for 10 years of Cycle 25 is estimated. The results are discussed by calculating the mean and smoothed values. The predicted Solar Cycle 25 shows features of a new Dalton Minimum together with Cycle 24. We conclude that Cycle 25 will be the marker of a new Dalton Minimum period.