Comparison of different predictive models and their effectiveness in sunspot number prediction

Abstract

Abstract Human activities and health are significantly influenced by solar activity. The sunspot number is one of the most commonly used measures of solar activity. The solar cycle’s quasi-periodic nature makes it an excellent choice for time series forecasting. Four models include three singular models, consisting of Long Short-Term Memory (LSTM), AutoRegressive Integrated Moving Average (ARIMA), and Seasonal AutoRegressive Integrated Moving Average (SARIMA), as well as a hybrid model were implemented to forecast maximum sunspot number of cycles 25 and 26. The hyperparameters of the singular models were optimized using Bayesian optimization. The LSTM-ARIMA hybrid model was able to achieve the best performance. The outstanding results of the LSTM-ARIMA model shows the potential of hybrid methods in improving the overall performance. Moreover, the LSTM model was able to outperform the ARIMA model, which demonstrates the ability of LSTM networks in learning from time-series data. The final model forecasts a peak sunspot number of 137.04 for Solar Cycle 25 in September 2024 and 164.3 for Solar Cycle 26 in December 2034, which is comparable to the National Aeronautics and Space Administration’s (NASA) prediction of 134.4 in October 2024 and 161.2 in December 2034.