Search

Efficient forecasting of solar flares is of significant importance for better risk prevention. Currently, there is relatively rare research on multi/four-classification of flares, and the influence of the number of time steps and data feature dimensions on the prediction performance of multi-class models has not been considered. In this study, we utilize the Space-weather HMI Active Region Patch (SHARP) data to develop two categories of models for multiclass flare prediction within 24 hr, including direct output four-classification models and four-classification models using a cascading scheme. The former encompasses Random Forest (RF) model, Long Short-Term Memory (LSTM) model, and Bidirectional LSTM (BLSTM) model, while the latter includes BLSTM Cascade (BLSTM-C) model and BLSTM Cascade with Attention Mechanism (BLSTM-C-A) model. These two categories of models are employed to contrast the impact of different numbers of time steps and the predictive performance in solar flare multi/four-classification. Additionally, we conduct, for the first time, feature importance analysis for multi/four-classification solar flare prediction using deep learning models. The main results are as follows: (1) As the number of time steps increases, the True Skill Statistic (TSS) scores of the four deep learning models improve, showing an overall upward trend in predictive performance. The models achieve their optimal performance when the number of time steps reaches 120. (2) Among the direct output four-class models, deep learning models (LSTM and BLSTM) outperform traditional machine learning model (RF). In both multi-class and binary-class predictions using deep learning, the BLSTM-C model performs better than other deep learning models (LSTM, BLSTM, and BLSTM-C-A). (3) In the feature importance analysis, the top-ranked important features include SAVNCPP and R_VALUE, while the least important features include SHRGT45 and MEANPOT. [ABSTRACT FROM AUTHOR]

The research on solar flare predicting holds significant practical and scientific value for safeguarding human activities. Current solar flare prediction models have not fully considered important factors such as time step length, nor have they conducted a comparative analysis of the physical features in multiple models or explored the consistency in the importance of features. In this work, based on SHARP data from SDO, we build 9 machine learning-based solar flare prediction models for binary "Yes" or "No" class prediction within the next 24 hours, and study the impact of different time steps and other factors on the forecasting performance. The main results are as follows. (1) The predictive performance of eight deep learning models shows an increasing trend as the time step length increases, and the models perform the best at the length of 40. (2) In predicting solar flares of ≥C class and ≥M class, the True Skill Statistic(TSS) of deep learning models consistently outperforms that of baseline model. For the same model, the TSS for predicting ≥M class flares generally exceeds that for predicting ≥C class flares. (3) The Brier Skill Score (BSS) of deep learning models significantly surpasses that of baseline model in predicting ≥C class flares. However, the BSS scores of the nine models are comparable for predicting ≥M class flares. For the same model, the BSS for predicting ≥C class flares is generally higher than that for predicting ≥M class flares. (4) Through feature importance analysis of multiple models, the common features that consistently rank at the top and bottom are identified. [ABSTRACT FROM AUTHOR]

Solar radio flux is an important indicator of solar activity and solar UV burst. Accurate prediction of solar radio flux plays a crucial role in preventing and mitigating the impact of solar activity on human productivity. We propose a novel approach for the first time to predict short-term radio flux trends using a bidirectional long short-term memory (BLSTM) network. This approach aims to address the unique characteristics of temporality and nonlinearity observed in solar radio flux data. Our model takes into account various frequency characteristics that impact radio flux. This allows it to learn temporal patterns within the data, ultimately enabling accurate predictions of radio flux for the next 30 minutes. The proposed method is experimentally applied to the radio flux dataset of the US Radio Solar Telescope Network (RSTN). The results show that, in most frequency bands, the BLSTM model exhibits superior prediction accuracy and greater sensitivity to peak responses compared to the LSTM model, LSTM-Attention (LSTM-A) model, BLSTM-Attention (BLSTM-A) model, and persistence model (PM). Consequently, the BLSTM model is better equipped to accurately forecast changes in radio flux for the next 30 minutes. [ABSTRACT FROM AUTHOR]

In this paper, we develop five machine learning models, neural network (NN), long short-term memory (LSTM), LSTM based on attention mechanism (LSTM-A), bidirectional LSTM (BLSTM), and BLSTM based on attention mechanism (BLSTM-A), for predicting whether a ≥C class or ≥M class flare will occur in an active region in the next 24 hr. We use the data base provided by the Space-weather Helioseismic and Magnetic Imager Active Region Patches of Solar Dynamic Observatory, including 10 magnetic field features of active regions from 2010 May 1 to 2018 September 13. The samples are labeled flare information (i.e. No-flare/C/M/X) using solar flare events catalogue provided by the Geostationary Operational Environmental Satellite and Solar Geophysical Data solar event reports. In addition, we generated 10 cross-validation sets from these data using the cross-validation method. Then, after training, validating, and testing our models, we compare the results with the true skill statistics (TSS) as the assessment metric. The main results are as follows. (1) The TSS scores for ≥C class are 0.5472 ± 0.0809, 0.6425 ± 0.0685, 0.6904 ± 0.0575, 0.6681 ± 0.0573, and 0.6833 ± 0.0531 for NN, LSTM, LSTM-A, BLSTM and BLSTM-A, respectively. The TSS scores for ≥M class are 0.5723 ± 0.1139, 0.6579 ± 0.0758, 0.5943 ± 0.0712, 0.6493 ± 0.0826, and 0.5932 ± 0.0723, respectively. (2) For the first time, we add an attention mechanism to BLSTM for flare prediction, which improves the performance of the model for ≥C class. (3) Among the five models, the prediction model based on deep learning algorithms is generally superior to the model based on the traditional machine learning algorithm. The performance of the LSTM models is comparable to that of the BLSTM models. In general, LSTM-A for ≥C class performs better than other models. In addition, we also discuss the influence of 10 features on LSTM-A, and we find that removing the least significant feature will result in better performance than using all 10 features together, and the TSS score of the model will improve to 0.7059 ± 0.0440. [ABSTRACT FROM AUTHOR]