Author(s)

Shubin Zhang

Corresponding Author

Shubin Zhang

Abstract

This paper presents an improved sunspot prediction model based on the Bidirectional Long Short-Term Memory (BiLSTM) network, which integrates solar physics parameters (e.g., the AP index) with historical sunspot data. The study reveals a significant correlation between the AP index and sunspot numbers, both exhibiting 11-year cyclic fluctuations, with the AP index peak lagging behind the sunspot number peak. Utilizing BiLSTM's ability to capture bidirectional time series features, the model considers both past and future data for predictions. The results demonstrate that the BiLSTM model significantly outperforms traditional LSTM in terms of prediction accuracy, as evidenced by metrics such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE). Specifically, the model predicts the peak number of sunspots and their occurrence time for the next solar activity cycle, indicating a peak of 185.4558 in July 2024 and 204.0451 in March 2036. This study provides a reliable tool for accurate solar activity cycle prediction, which is crucial for space weather research and applications.

Keywords

Sunspot Prediction; BiLSTM; AP Index; Solar Activity Cycle