Short-term Prediction Of Flare Exponentially Smoothed Values Based On LSTM

Solar flare is a severe form of solar activity.The enhancement of X-ray caused by a strong solar flare will lead to shortwave radio attenuation,which will affect the safety of radio communication system,global positioning system,satellite and astronauts.These hazards will cause a lot of economic and commercial losses.Therefore,the establishment of solar flare prediction model is of great significance for space weather forecast.The flare index is a quantitative index to describe the intensity of solar flare activity.It is one of the most important solar activity indexes in the field of solar radiation research.Compared with other solar activity index series,the flare index time series is more abrupt and difficult to predict.The smooth value of flare index can slow down the fluctuation and keep the overall trend of flare index.We can predict the overall activity level of solar flares in the next week by predicting the smooth value of flare index.In this paper,the Hurst index values of flare index and flare index smoothing value are calculated by R/S analysis method,which shows the predictability of flare index smoothing value,and the predictability of flare index smoothing value is better than flare index,which provides a basis for subsequent experiments.Then,the long-term and short-term memory network model suitable for processing time series is selected to extract the time series information from the flare index,and the prediction problem is transformed into the supervision problem in machine learning by sliding window method to predict the smooth value of the solar flare index in the next week.Aiming at the problem that the smooth value of the flare index needs to use the future information,a new method based on Karl Fischer is proposed The prediction model of flare exponential smoothing value based on Mann filter and long-term and short-term memory network.Firstly,the optimal super parameters of the flare exponential smoothing value prediction network model are determined,and the prediction results from the first day to the seventh day are extracted from the prediction results for error analysis.The model prediction results are quantitatively analyzed by the average absolute error and other evaluation indexes.The experimental results show the superiority of the proposed prediction model in predicting the flare exponential smoothing value.Compared with LSTM neural network and RNN neural network,the model proposed in this paper has higher accuracy.Finally,the model proposed in this paper is combined with the springboot framework to establish the flare exponential smoothing value prediction system.