Predicting Solar cycle 25 using an optimized long short-term memory model based on sunspot area data

Abstract

In this paper, an optimized long short-term memory (LSTM) model was proposed to deal with the monthly sunspot area (SSA) data, aiming to predict the peak amplitude of SSA and the occurring time for Solar Cycle 25, and obtain the forecasting maximum value of sunspot number (SSN) and the reaching time to maximum. The “re-prediction” process in LSTM + was employed with the latest prediction results obtained from the previous prediction as the input for the next prediction calculation. The accuracy and reliability of LSTM + were performed by the validation of the predicted peak amplitude and occurring time of Solar Cycles 21 to 24. The prediction error between the predicted and the observed peak amplitude for cycles 21 to 24 were 8.85 %, 4.49 %, 2.88 %, and 4.57 %, respectively, and the error for the occurring time for 4 cycles was all within 6 months. The predicted peak amplitude of SSA for Solar Cycle 25 was obtained as 2562.5 with LSTM+, and the maximum value of SSN was calculated as 213 based on the relation between SSA and SSN, which would be stronger than that of Solar Cycle 24. And according to the prediction result, Solar Cycle 25 would reach the peak around January 2025.