Forecast of the Solar 10.7-cm Flux based on Neural Networks with Attention Mechanism

The upper atmosphere of the Earth has a significant impact on various space missions. Research has found that the value of the solar 10.7-cm flux is closely related to the density of the Earth's upper atmosphere. Therefore, in many atmospheric models, the solar 10.7-cm flux is an indispensable input parameter. This paper studies the short-term prediction of solar 10.7-cm flux based on neural networks, and in particular, studies the application of attention mechanism in short-term prediction of solar 10.7-cm flux. By using attention mechanism, the model can be forced to pay more attention to the hidden feature information that has a greater impact on the current prediction result when predicting the solar 10.7-cm flux for the next three days. Experimental results show that using the model proposed in this paper can effectively predict the short-term solar 10.7-cm flux. When attention mechanism is not used, the mean absolute percentage error of one-day-ahead prediction is 2.34%, and the root mean square error is 2.90sfu. When attention mechanism is used, the mean absolute percentage error of one-day-ahead prediction is 2.04%, and the root mean square error is 2.67sfu. Compared with the model without attention mechanism, the model using attention mechanism has higher prediction accuracy, verifying the effectiveness of attention mechanism in short-term prediction of solar 10.7-cm flux.