Deep Flare Net (DeFN) Model for Solar Flare Prediction

Abstract

We developed a solar flare prediction model using a deep neural network (DNN), named Deep Flare Net (DeFN). The model can calculate the probability of flares occurring in the following 24 h in each active region, which is used to determine the most likely maximum classes of flares via a binary classification (e.g., ≥M class versus <M class or ≥C class versus <C class). From 3×10 5 observation images taken during 2010-2015 by Solar Dynamic Observatory, we automatically detected sunspots and calculated 79 features for each region, to which flare occurrence labels of X-, M-, and C-class were attached. We adopted the features used in and added some features for operational prediction: coronal hot brightening at 131Å (T≥10 7 K) and the X-ray and 131Å intensity data 1 and 2 h before an image. For operational evaluation, we divided the database into two for training and testing: the dataset in 2010-2014 for training and the one in 2015 for testing. The DeFN model consists of deep multilayer neural networks, formed by adapting skip connections and batch normalizations. To statistically predict flares, the DeFN model was trained to optimize the skill score, i.e., the true skill statistic (TSS). As a result, we succeeded in predicting flares with TSS=0.80 for ≥M-class flares and TSS=0.63 for ≥C-class flares. Note that in usual DNN models, the prediction process is a black box. However, in the DeFN model, the features are manually selected, and it is possible to analyze which features are effective for prediction after evaluation.

Predicting Solar Flares Using a Long Short-term Memory Network

“…Following Bobra & Couvidat (2015), Jonas et al (2018) and Nishizuka et al (2018), we intend to use past observations of an AR to predict its future flaring activity. Specifically, we want to solve the following binary classification problem: will this AR produce a Υ-class flare within the next 24 hours?…”

“…We next compare our LSTM framework with five closely related machine learning methods including multilayer perceptrons (MLP) (Haykin & Network 2004;Florios et al 2018), Jordan network (JN) (Jordan 1997), support vector machines (SVM) (Qahwaji & Colak 2007;Yuan et al 2010;Bobra & Couvidat 2015;Boucheron et al 2015;Muranushi et al 2015;Florios et al 2018), random forests (RF) (Barnes et al 2016;Liu et al 2017;Florios et al 2018), and a recently published deep learning-based method, Deep Flare Net (DeFN; Nishizuka et al 2018). All these methods including ours (LSTM) can be used as a binary classification model (Nishizuka et al 2018;Jonas et al 2018) or a probabilistic forecasting model (Florios et al 2018). A binary classification model predicts whether or not an AR will produce a ≥M5.0-(≥M-, ≥C-, respectively) class flare within the next 24 hours.…”

“…Commonly used machine learning methods for flare prediction include decision trees (Yu et al 2009(Yu et al , 2010, random forests (Barnes et al 2016;Liu et al 2017;Florios et al 2018;Breiman 2001), k-nearest neighbors Huang et al 2013;Winter & Balasubramaniam 2015;Nishizuka et al 2017), support vector machines (Qahwaji & Colak 2007;Yuan et al 2010;Bobra & Couvidat 2015;Boucheron et al 2015;Muranushi et al 2015;Florios et al 2018), ordinal logistic regression (Song et al 2009), the least absolute shrinkage and selection operator (LASSO) (Benvenuto et al 2018;Jonas et al 2018), extremely randomized trees (Nishizuka et al 2017), and neural networks (Qahwaji & Colak 2007;Wang et al 2008;Colak & Qahwaji 2009;Higgins et al 2011;Ahmed et al 2013). Recently, Nishizuka et al (2018) adopted a deep neural network, named Deep Flare Net, for flare prediction.…”

Leveraging the mathematics of shape for solar magnetic eruption prediction

“…By choosing w no− f lare : w f lare = 1 : 100, we over-penalize the minority class (flaring magnetograms) to offset the effect of its size. This is one of the many ways to mitigate imbalance in training sets, and has been used in the flare-prediction literature (Bobra and Couvidat, 2015;Nishizuka et al, 2018). Finally, we use the Adagrad optimizer (Duchi et al, 2011) to perform the weight update during the backpropagation phase.…”

“…Nishizuka et al (2017) have also applied decision trees and clustering to the same task. Neural networks, which go beyond simple binary classification by learning complex nonlinear relationships among their inputs, have also been used to great advantage by Nishizuka et al (2018). A variety of other ML algorithms, such as Bayesian networks (Yu et al, 2010), radial basis model networks (Colak and Qahwaji, 2009), logistic regression (Yuan et al, 2010), LASSO regression (Campi et al, 2019), and random forests or ERTs (Nishizuka et al, 2017;Campi et al, 2019) have also been implemented for solar flare prediction with some degree of success.…”