Research on Medium-Term Forecast of the Solar Radiation Flux F10.7 Index

The solar radiation flux F10.7 index is an important parameter reflecting the level of solar activity. The accuracy of its medium-term forecast plays a decisive role in predicting monthly trends of atmospheric density in the thermosphere, which has considerable application value in key space missions such as spacecraft reentry and collision avoidance. Given the excellent performance of neural networks in handling time-series prediction problems, this study focuses on the medium-term prediction method of the F10.7 index based on neural networks. This study employs Long Short-Term Memory (LSTM) neural networks as the main model, using observed F10.7 index data from the past five solar rotation cycles to predict the F10.7 index for the next 27 days. Simulation results show that, on the test set, the mean absolute percentage error (MAPE) between the predicted and observed values 27 days in advance is less than 6%, with a root mean square error (RMSE) within 8.5 sfu. This provides strong evidence for the effectiveness of this model in medium-term F10.7 index prediction. It is worth noting that during high solar activity years, the F10.7 index prediction errors on both the training and test sets tend to be larger. Additionally, increasing the dimension of LSTM's hidden layers does not necessarily improve model performance. This finding reveals that there is no linear relationship between model performance and complexity. This study provides strong support for improving the accuracy of medium-term F10.7 index prediction and provides more accurate predictive insights for related space missions. Future research can further explore model parameter optimization and multi-factor fusion methods to enhance the prediction accuracy of the F10.7 index.