Forecasting Solar Cycle Using the Time-series Dense Encoder Deep Learning Model

The solar cycle is a phenomenon by the Sun’s quasi-periodic regular activities and occurring approximately every 11 years. Intense solar activities can disrupt the ionosphere, affect communication and navigation systems. Hence, accurately predicting the intensity of the solar cycle holds great significance. However, when it comes to predicting the solar cycle, which involves a long-term time series, many existing time-series forecasting methods have fallen short in terms of accuracy and efficiency. The Time-series Dense Encoder (TiDE) model was proposed as a deep learning solution tailored for long time series prediction, which was founded upon a multi-layer perceptron structure, outperformed the current state-of-the-art models in accuracy but also boasted an training efficiency. In this study, we proposed a method based on TiDE for forecasting solar cycle. By utilizing the trained TiDE model, we predicted the test set from solar cycles 19 to 25. The average Mean Absolute Percentage Error was 32.02, Root Mean Square Error was 30.3, Mean Absolute Error was 23.32 and R2_SCORE was 0.76, outperforming the traditional N-BEATS, CNN, RNN, LSTM and GRU deep learning models in terms of accuracy and training efficiency. Subsequently, the model was applied to predict the peak of solar cycle 25 and 26. For solar cyle 25, the peak time has ended, followed by solar cycle 26’s stronger peak of 199.3, vary between in 170.8-221.9, projected to occur on April 2034.