基于深度学习的太阳F_10.7辐射通量的短期预报研究

F_10.7太阳辐射通量作为输入参数被广泛运用于大气经验模型、电离层模型等空间环境模型,其预报精度直接影响航天器轨道预报精度.采用时间序列法统计了太阳辐射通量F10.7指数和太阳黑子数(SSN)的关系,给出了两者之间的线性关系,在此基础上提出了一种基于长短时记忆神经网络(Long and Short Term Memory,LSTM)的预报方法,方法结合了54 d太阳辐射通量指数和SSN历史数据来对F_10.7进行未来7 d短期预报,并与其他预报方法的预报结果进行了比较,结果表明:(1)所建短期预报7 d方法模型的性能优于美国空间天气预报中心(Space Weather Prediction Center, SWPC)的方法,预测值和观测值的相关系数(CC)达到0.96,同时其均方根误差约为11.62个太阳辐射通量单位(sfu),预报结果的均方根误差(RMSE)低于SWPC,下降约11%;(2)对预测的23、24周太阳活动年结果统计表明,太阳活动高年的第7 d F10.7指数预报平均绝对百分比误差(MAPE)最优可达12.9%以内,低年最优可达2...

Short-Term Prediction of Solar F_10.7 Radiation Flux Based on Deep Learning

Solar radiation flux F_10.7 is widely used in space environment models such as the empirical model of atmosphere and the ionospheric model as an input parameter, and its prediction accuracy directly affects the precision of spacecraft orbit prediction. The relationship between solar radiation flux (F_{10.7) and sunspot number (SSN) is calculated by the time series method, and the linear relationship between them is given. On this basis, we proposed a Short-Term prediction method for F_10.7 in the next 7 days based on the 54-day solar radiation flux index combined with the sunspots number into the LSTM (Long and Short Term Memory) neural network, compared the prediction results with those of other forecasting methods. The results show: (1) The performance of the proposed 7-day method model is better than that of the Space Weather Prediction Center (SWPC), and the correlation coefficient (CC) between the predicted and observed values reaches 0.96. At the same time, the root mean square error is about 11.62 solar radiation flux units, and the RMSE of the forecast result is lower than that of SWPC, which decreases by about 11%. (2) According to the statistical results of 23 and 24 solar activity cycle year, the optimal MAPE (Mean Absolute Percentage Error) of the average absolute percentage error of F_10.7 index on the seventh day of high solar activity year can reach 12.9%, and the optimal MAPE of low solar activity year can reach 2.5%. (3) The results of combined sunspot number LSTM and LSTM using only F_10.7 showed that the newly introduced sunspot number was effective in improving the prediction of LSTM, and the root mean square error (RMSE) of these two models were lower than that of SWPC by about 11% and 5%, respectively.