Research On Solar Flare Forecast Model Based On Deep Learning

Solar flares are an important phenomenon of solar eruptive activity,an important warning device of space weather disturbance,and an important forecast content of space weather forecasting.For flare forecasting,the evolution information of the solar active area cannot be ignored.In recent years,deep learning technology has continued to develop.Convolutional neural networks in deep learning methods can automatically extract image features,and Recurrent neural network can extract time series information of samples.At present,deep learning has been widely used in speech recognition,natural language processing,and target Fields such as detection,identification and classification.This paper uses deep learning methods to separately propose a solar flare forecast model and a solar flare recommendation model.The solar flare prediction model uses long-and short-term memory networks to analyze the time-series evolution of solar active areas,so as to predict the occurrence of flares in the next 48 hours.In this paper,a flare data set is established by means of a sliding window,and the importance of each physical parameter in the SDO/HMI SHARP data is analyzed by the XGBoost method during the modeling process,and a long-and short-term memory neural network is used to establish a flare prediction model.The model is better than traditional machine learning models in terms of accuracy rate and real skill parameters,which are 0.7483 and 0.7402,respectively.The false reporting rate and accuracy rate are similar to traditional machine learning models,which are 0.0081 and 0.9894 respectively,and the overall effect of the model is better than traditional machine learning models.In the recommendation model based on deep learning,the pre-trained ResNet50 image feature extractor and the long short-term memory neural network time series feature extractor are used to extract the features of the samples,and finally the Euclidean distance algorithm and the cosine similarity algorithm are used to calculate the sample and historical database.The similarity of the sample,thus recommending events similar to the sample,and giving an early warning of the event.This paper establishes a solar flare recommendation model,which can recommend solar flare events similar to the current sample in history,and forecast the occurrence of solar flare events in the next 6 hours,12 hours,24 hours,and 48 hours.Through model evaluation,a recommendation model based on the cosine similarity algorithm and the next 48-hour time sequence feature extractor was finally selected.The recommended effects of the model are recall rate 0.8410,accuracy rate 0.8262,accuracy rate 0.9472 and F1-Score 0.8335.The model is in It performs well in short-term forecasting.In addition to warning of flare events,historical similar conditions can also be recommended for forecasters’ reference.The application of the recommended algorithm to solar flare prediction has research significance and practical value.